(ICPR10)Multiple Human Tracking Based on Multi-View Upper-Body Detection and Discriminative Learning

合集下载

人工智能导论考核试卷

2.监督学习：基于标记数据学习；无监督学习：从无标记数据中发现模式；强化学习：通过奖励与惩罚学习策略。案例：监督学习用于邮件分类，无监督学习用于客户细分，强化学习用于游戏AI。
3. CNN通过卷积和池化操作提取图像特征，降低参数数量，提高模型泛化能力，从而提高图像识别准确性。
4.伦理问题：隐私保护、算法偏见、责任归属。解决策略：制定伦理准则、透明度提升、多样化团队、责任追溯机制。
1.人工智能包括以下哪些技术领域？()
A.机器学习
B.语音识别
C.量子计算
D.数据挖掘
E.虚拟现实
2.以下哪些属于监督学习算法？()
A.支持向量机
B.决策树
C. K-均值聚类
D.线性回归
E.随机森林
3.深度学习中的卷积神经网络（CNN）主要用于哪些任务？()
A.图像分类
B.语音识别
C.自然语言处理
D.视频分析
人工智能导论考核试卷
考生姓名：__________答题日期：__________得分：__________判卷人：__________
一、单项选择题（本题共20小题，每小题1分，共20分，在每小题给出的四个选项中，只有一项是符合题目要求的）
1.以下哪个不是人工智能的研究领域？()
A.机器学习
B.深度学习
D.随机森林
E.支持向量回归
9.以下哪些是推荐系统中的冷启动问题？()
A.用户冷启动
B.项目冷启动
C.模型冷启动
D.数据冷启动
E.系统冷启动
10.以下哪些是迁移学习的主要挑战？()
A.数据分布差异
B.标签空间不匹配
C.模型泛化能力不足
D.源域数据不足
E.目标域数据过拟合

!How Far are We from Solving Pedestrian Detection

from Solving Pedestrian Detection?Mohamed Omran,Jan Hosang,and Bernt SchielePlanck Institute for Informatics Saarbrücken,Germanystname@mpi-inf.mpg.deAbstractEncouraged by the recent progress in pedestrian detec-tion,we investigate the gap between current state-of-the-art methods and the “perfect single frame detector”.We en-able our analysis by creating a human baseline for pedes-trian detection (over the Caltech dataset),and by manually clustering the recurrent errors of a top detector.Our res-ults characterize both localization and background-versus-foreground errors.To address localization errors we study the impact of training annotation noise on the detector performance,and show that we can improve even with a small portion of sanitized training data.To address background/foreground discrimination,we study convnets for pedestrian detection,and discuss which factors affect their performance.Other than our in-depth analysis,we report top perform-ance on the Caltech dataset,and provide a new sanitized set of training and test annotations 1.1.IntroductionObject detection has received great attention during re-cent years.Pedestrian detection is a canonical sub-problem that remains a popular topic of research due to its diverse applications.Despite the extensive research on pedestrian detection,recent papers still show signiﬁcant improvements,suggest-ing that a saturation point has not yet been reached.In this paper we analyse the gap between the state of the art and a newly created human baseline (section 3.1).The results indicate that there is still a ten fold improvement to be made before reaching human performance.We aim to investigate which factors will help close this gap.We analyse failure cases of top performing pedestrian detectors and diagnose what should be changed to further push performance.We show several different analysis,in-cluding human inspection,automated analysis of problem1Ifyou are interested in our new annotations,please contact Shanshan Zhang.1010101010Figure 1:Overview of the top results on the Caltech-USA pedestrian benchmark (CVPR2015snapshot).At ∼95%recall,state-of-the-art detectors make ten times more errors than the human baseline.cases (e.g.blur,contrast),and oracle experiments (section 3.2).Our results indicate that localization is an important source of high conﬁdence false positives.We address this aspect by improving the training set alignment quality,both by manually sanitising the Caltech training annotations and via algorithmic means for the remaining training samples (sections 3.3and 4.1).To address background versus foreground discrimina-tion,we study convnets for pedestrian detection,and dis-cuss which factors affect their performance (section 4.2).1.1.Related workIn the last years,diverse efforts have been made to im-prove the performance of pedestrian detection.Following the success of integral channel feature detector (ICF)[6,5],many variants [22,23,16,18]were proposed and showed signiﬁcant improvement.A recent review of pedestrian de-tection [3]concludes that improved features have been driv-ing performance and are likely to continue doing so.It also shows that optical ﬂow [19]and context information [17]are complementary to image features and can further boost 1a r X i v :1602.01237v 1 [c s .C V ] 3 F eb 2016detection accuracy.Byﬁne-tuning a model pre-trained on external data convolution neural networks(convnets)have also reached state-of-the-art performance[15,20].Most of the recent papers focus on introducing novelty and better results,but neglect the analysis of the resulting system.Some analysis work can be found for general ob-ject detection[1,14];in contrast,in theﬁeld of pedestrian detection,this kind of analysis is rarely done.In2008,[21] provided a failure analysis on the INRIA dataset,which is relatively small.The best method considered in the2012 Caltech dataset survey[7]had10×more false positives at20%recall than the methods considered here,and no method had reached the95%mark.Since pedestrian detection has improved signiﬁcantly in recent years,a deeper and more comprehensive analysis based on state-of-the-art detectors is valuable to provide better understanding as to where future efforts would best be invested.1.2.ContributionsOur key contributions are as follows:(a)We provide a detailed analysis of a state-of-the-art ped-estrian detection system,providing insights into failure cases.(b)We provide a human baseline for the Caltech Pedestrian Benchmark;as well as a sanitised version of the annotations to serve as new,high quality ground truth for the training and test sets of the benchmark.The data will be public. (c)We analyse how much the quality of training data affects the detector.More speciﬁcally we quantify how much bet-ter alignment and fewer annotation mistakes can improve performance.(d)Using the insights of the analysis,we explore variants of top performing methods:ﬁltered channel feature detector [23]and R-CNN detector[13,15],and show improvements over the baselines.2.PreliminariesBefore delving into our analysis,let us describe the data-sets in use,their metrics,and our baseline detector.2.1.Caltech-USA pedestrian detection benchmarkAmongst existing pedestrian datasets[4,9,8],KITTI [11]and Caltech-USA are currently the most popular ones. In this work we focus on the Caltech-USA benchmark[7] which consists of2.5hours of30Hz video recorded from a vehicle traversing the streets of Los Angeles,USA.The video annotations amount to a total of350000bound-ing boxes covering∼2300unique pedestrians.Detec-tion methods are evaluated on a test set consisting of4024 frames.The provided evaluation toolbox generates plotsFilter type MR O−2ACF[5]44.2SCF[3]34.8LDCF[16]24.8RotatedFilters19.2Checkerboards18.5Table1:Theﬁltertype determines theICF methods quality.Base detector MR O−2+Context+FlowOrig.2Ped[17]48~5pp/Orig.SDt[19]45/8ppSCF[3]355pp4ppCheckerboards19~01ppTable2:Detection quality gain ofadding context[17]and opticalﬂow[19],as function of the base detector.for different subsets of the test set based on annotation size, occlusion level and aspect ratio.The established proced-ure for training is to use every30th video frame which res-ults in a total of4250frames with∼1600pedestrian cut-outs.More recently,methods which can leverage more data for training have resorted to aﬁner sampling of the videos [16,23],yielding up to10×as much data for training than the standard“1×”setting.MR O,MR N In the standard Caltech evaluation[7]the miss rate(MR)is averaged over the low precision range of [10−2,100]FPPI.This metric does not reﬂect well improve-ments in localization errors(lowest FPPI range).Aiming for a more complete evaluation,we extend the evaluation FPPI range from traditional[10−2,100]to[10−4,100],we denote these MR O−2and MR O−4.O stands for“original an-notations”.In section3.3we introduce new annotations, and mark evaluations done there as MR N−2and MR N−4.We expect the MR−4metric to become more important as de-tectors get stronger.2.2.Filtered channel features detectorFor the analysis in this paper we consider all methods published on the Caltech Pedestrian benchmark,up to the last major conference(CVPR2015).As shown inﬁgure1, the best method at the time is Checkerboards,and most of the top performing methods are of its same family.The Checkerboards detector[23]is a generalization of the Integral Channels Feature detector(ICF)[6],which ﬁlters the HOG+LUV feature channels before feeding them into a boosted decision forest.We compare the performance of several detectors from the ICF family in table1,where we can see a big improve-ment from44.2%to18.5%MR O−2by introducingﬁlters over the feature channels and optimizing theﬁlter bank.Current top performing convnets methods[15,20]are sensitive to the underlying detection proposals,thus weﬁrst focus on the proposals by optimizing theﬁltered channel feature detectors(more on convnets in section4.2). Rotatedﬁlters For the experiments involving train-ing new models(in section 4.1)we use our own re-implementation of Checkerboards[23],based on the LDCF[16]codebase.To improve the training time we decrease the number ofﬁlters from61in the originalCheckerboards down to9ﬁlters.Our so-called Rota-tedFilters are a simpliﬁed version of LDCF,applied at three different scales(in the same spirit as Squares-ChnFtrs(SCF)[3]).More details on theﬁlters are given in the supplementary material.As shown in table1,Ro-tatedFilters are signiﬁcantly better than the original LDCF,and only1pp(percent point)worse than Checker-boards,yet run6×faster at train and test time. Additional cues The review[3]showed that context and opticalﬂow information can help improve detections. However,as the detector quality improves(table1)the re-turns obtained from these additional cues erodes(table2). Without re-engineering such cues,gains in detection must come from the core detector.3.Analysing the state of the artIn this section we estimate a lower bound on the re-maining progress available,analyse the mistakes of current pedestrian detectors,and propose new annotations to better measure future progress.3.1.Are we reaching saturation?Progress on pedestrian detection has been showing no sign of slowing in recent years[23,20,3],despite recent im-pressive gains in performance.How much progress can still be expected on current benchmarks?To answer this ques-tion,we propose to use a human baseline as lower bound. We asked domain experts to manually“detect”pedestrians in the Caltech-USA test set;machine detection algorithms should be able to at least reach human performance and, eventually,superhuman performance.Human baseline protocol To ensure a fair comparison with existing detectors,we focus on the single frame mon-ocular detection setting.Frames are presented to annotators in random order,and without access to surrounding frames from the source videos.Annotators have to rely on pedes-trian appearance and single-frame context rather than(long-term)motion cues.The Caltech benchmark normalizes the aspect ratio of all detection boxes[7].Thus our human annotations are done by drawing a line from the top of the head to the point between both feet.A bounding box is then automatically generated such that its centre coincides with the centre point of the manually-drawn axis,see illustration inﬁgure2.This procedure ensures the box is well centred on the subject (which is hard to achieve when marking a bounding box).To check for consistency among the two annotators,we produced duplicate annotations for a subset of the test im-ages(∼10%),and evaluated these separately.With a Intersection over Union(IoU)≥0.5matching criterion, the results were identical up to a single boundingbox.Figure2:Illustration of bounding box generation for human baseline.The annotator only needs to draw a line from the top of the head to the central point between both feet,a tight bounding box is then automatically generated. Conclusion Inﬁgure3,we compare our human baseline with other top performing methods on different subsets of the test data(varying height ranges and occlu-sion levels).Weﬁnd that the human baseline widely out-performs state-of-the-art detectors in all settings2,indicat-ing that there is still room for improvement for automatic methods.3.2.Failure analysisSince there is room to grow for existing detectors,one might want to know:when do they fail?In this section we analyse detection mistakes of Checkerboards,which obtains top performance on most subsets of the test set(see ﬁgure3).Since most top methods ofﬁgure1are of the ICF family,we expect a similar behaviour for them too.Meth-ods using convnets with proposals based on ICF detectors will also be affected.3.2.1Error sourcesThere are two types of errors a detector can do:false pos-itives(detections on background or poorly localized detec-tions)and false negatives(low-scoring or missing pedes-trian detections).In this analysis,we look into false positive and false negative detections at0.1false positives per im-age(FPPI,1false positive every10images),and manually cluster them(one to one mapping)into visually distinctive groups.A total of402false positive and148false negative detections(missing recall)are categorized by error type. False positives After inspection,we end up having all false positives clustered in eleven categories,shown inﬁg-ure4a.These categories fall into three groups:localization, background,and annotation errors.Background errors are the most common ones,mainly ver-tical structures(e.g.ﬁgure5b),tree leaves,and trafﬁc lights. This indicates that the detectors need to be extended with a better vertical context,providing visibility over larger struc-tures and a rough height estimate.Localization errors are dominated by double detections2Except for IoU≥0.8.This is due to issues with the ground truth, discussed in section3.3.Reasonable (IoU >= 0.5)Height > 80Height in [50,80]Height in [30,50]020406080100HumanBaselineCheckerboards RotatedFiltersm i s s r a t eFigure 3:Detection quality (log-average miss rate)for different test set subsets.Each group shows the human baseline,the Checkerboards [23]and RotatedFilters detectors,as well as the next top three (unspeciﬁed)methods (different for each setting).The corresponding curves are provided in the supplementary material.(high scoring detections covering the same pedestrian,e.g.ﬁgure 5a ).This indicates that improved detectors need to have more localized responses (peakier score maps)and/or a different non-maxima suppression strategy.In sections 3.3and 4.1we explore how to improve the detector localiz-ation.The annotation errors are mainly missing ignore regions,and a few missing person annotations.In section 3.3we revisit the Caltech annotations.False negatives Our clustering results in ﬁgure 4b show the well known difﬁculty of detecting small and oc-cluded objects.We hypothesise that low scoring side-view persons and cyclists may be due to a dataset bias,i.e.these cases are under-represented in the training set (most per-sons are non-cyclist walking on the side-walk,parallel to the car).Augmenting the training set with external images for these cases might be an effective strategy.To understand better the issue with small pedestrians,we measure size,blur,and contrast for each (true or false)de-tection.We observed that small persons are commonly sat-urated (over or under exposed)and blurry,and thus hypo-thesised that this might be an underlying factor for weak detection (other than simply having fewer pixels to make the decision).Our results indicate however that this is not the case.As ﬁgure 4c illustrates,there seems to be no cor-relation between low detection score and low contrast.This also holds for the blur case,detailed plots are in the sup-plementary material.We conclude that the small number of pixels is the true source of difﬁculty.Improving small objects detection thus need to rely on making proper use of all pixels available,both inside the window and in the surrounding context,as well as across time.Conclusion Our analysis shows that false positive er-rors have well deﬁned sources that can be speciﬁcally tar-geted with the strategies suggested above.A fraction of the false negatives are also addressable,albeit the small and oc-cluded pedestrians remain a (hard and)signiﬁcant problem.20406080100120# e r r o r s 0100200300loc a liz a tion ba c k g round a nnota e rrors#e r r o r s (a)False positive sources15304560# e r r o r s (b)False negative sources(c)Contrast versus detection scoreFigure 4:Errors analysis of Checkerboards [23]on the test set.(a)double detectionFigure 5:Example of analysed false positive cases (red box).Additional ones in supplementary material.3.2.2Oracle test casesThe analysis of section 3.2.1focused on errors counts.For area-under-the-curve metrics,such astheones used in Caltech,high-scoring errors matter more than low-scoring ones.In this section we directly measure the impact of loc-alization and background-vs-foreground errors on the de-tection quality metric (log-average miss-rate)by using or-acle test cases.In the oracle case for localization,all false positives that overlap with ground truth are ignored for evaluation.In the oracle tests for background-vs-foreground,all false posit-ives that do not overlap with ground truth are ignored.Figure 6a shows that ﬁxing localization mistakes im-proves performance in the low FPPI region;while ﬁxing background mistakes improves results in the high FPPI re-gion.Fixing both types of mistakes results zero errors,even though this is not immediately visible due to the double log plot.In ﬁgure 6b we show the gains to be obtained in MR O −4terms by ﬁxing localization or background issues.When comparing the eight top performing methods we ﬁnd that most methods would boost performance signiﬁcantly by ﬁx-ing either problem.Note that due to the log-log nature of the numbers,the sum of localization and background deltas do not add up to the total miss-rate.Conclusion For most top performing methods localiz-ation and background-vs-foreground errors have equal im-pact on the detection quality.They are equally important.3.3.Improved Caltech-USA annotationsWhen evaluating our human baseline (and other meth-ods)with a strict IoU ≥0.8we notice in ﬁgure 3that the performance drops.The original annotation protocol is based on interpolating sparse annotations across multiple frames [7],and these sparse annotations are not necessar-ily located on the evaluated frames.After close inspection we notice that this interpolation generates a systematic off-set in the annotations.Humans walk with a natural up and down oscillation that is not modelled by the linear interpol-ation used,thus in most frames have shifted bounding box annotations.This effect is not noticeable when using the forgiving IoU ≥0.5,however such noise in the annotations is a hurdle when aiming to improve object localization.1010−210−110010false positives per image18.47(33.20)% Checkerboards15.94(25.49)% Checkerboards (localization oracle)11.92(26.17)% Checkerboards (background oracle)(a)Original and two oracle curves for Checkerboards de-tector.Legend indicates MR O −2 MR O −4 .(b)Comparison of miss-rate gain (∆MR O −4)for top performing methods.Figure 6:Oracle cases evaluation over Caltech test set.Both localization and background-versus-foreground show important room for improvement.(a)False annotations (b)Poor alignmentFigure 7:Examples of errors in original annotations.New annotations in green,original ones in red.This localization issues together with the annotation er-rors detected in section 3.2.1motivated us to create a new set of improved annotations for the Caltech pedestrians dataset.Our aim is two fold;on one side we want to provide a more accurate evaluation of the state of the art,in particu-lar an evaluation suitable to close the “last 20%”of the prob-lem.On the other side,we want to have training annotations and evaluate how much improved annotations lead to better detections.We evaluate this second aspect in section 4.1.New annotation protocol Our human baseline focused on a fair comparison with single frame methods.Our new annotations are done both on the test and training 1×set,and focus on high quality.The annotators are allowed to look at the full video to decide if a person is present or not,they are request to mark ignore regions in areas cov-ering crowds,human shapes that are not persons (posters,statues,etc.),and in areas that could not be decided as cer-tainly not containing a person.Each person annotation is done by drawing a line from the top of the head to the point between both feet,the same as human baseline.The annot-ators must hallucinate head and feet if these are not visible. When the person is not fully visible,they must also annotate a rectangle around the largest visible region.This allows to estimate the occlusion level in a similar fashion as the ori-ginal Caltech annotations.The new annotations do share some bounding boxes with the human baseline(when no correction was needed),thus the human baseline cannot be used to do analysis across different IoU thresholds over the new test set.In summary,our new annotations differ from the human baseline in the following aspects:both training and test sets are annotated,ignore regions and occlusions are also an-notated,full video data is used for decision,and multiple revisions of the same image are allowed.After creating a full independent set of annotations,we con-solidated the new annotations by cross-validating with the old annotations.Any correct old annotation not accounted for in the new set,was added too.Our new annotations correct several types of errors in the existing annotations,such as misalignments(ﬁgure 7b),missing annotations(false negatives),false annotations (false positives,ﬁgure7a),and the inconsistent use of“ig-nore”regions.Our new annotations will be publicly avail-able.Additional examples of“original versus new annota-tions”provided in the supplementary material,as well as visualization software to inspect them frame by frame. Better alignment In table3we show quantitative evid-ence that our new annotations are at least more precisely localized than the original ones.We summarize the align-ment quality of a detector via the median IoU between true positive detections and a give set of annotations.When evaluating with the original annotations(“median IoU O”column in table3),only the model trained with original annotations has good localization.However,when evalu-ating with the new annotations(“median IoU N”column) both the model trained on INRIA data,and on the new an-notations reach high localization accuracy.This indicates that our new annotations are indeed better aligned,just as INRIA annotations are better aligned than Caltech.Detailed IoU curves for multiple detectors are provided in the supplementary material.Section4.1describes the RotatedFilters-New10×entry.4.Improving the state of the artIn this section we leverage the insights of the analysis, to improve localization and background-versus-foreground discrimination of our baseline detector.DetectorTrainingdataMedianIoU OMedianIoU N Roerei[2]INRIA0.760.84RotatedFilters Orig.10×0.800.77RotatedFilters New10×0.760.85 Table3:Median IoU of true positives for detectors trained on different data,evaluated on original and new Caltech test.Models trained on INRIA align well with our new an-notations,conﬁrming that they are more precise than previ-ous ones.Curves for other detectors in the supplement.Detector Anno.variant MR O−2MR N−2ACFOriginal36.9040.97Pruned36.4135.62New41.2934.33 RotatedFiltersOriginal28.6333.03Pruned23.8725.91New31.6525.74 Table4:Effects of different training annotations on detec-tion quality on validation set(1×training set).Italic num-bers have matching training and test sets.Both detectors im-prove on the original annotations,when using the“pruned”variant(see§4.1).4.1.Impact of training annotationsWith new annotations at hand we want to understand what is the impact of annotation quality on detection qual-ity.We will train ACF[5]and RotatedFilters mod-els(introduced in section2.2)using different training sets and evaluate on both original and new annotations(i.e. MR O−2,MR O−4and MR N−2,MR N−4).Note that both detect-ors are trained via boosting and thus inherently sensitive to annotation noise.Pruning beneﬁts Table4shows results when training with original,new and pruned annotations(using a5/6+1/6 training and validation split of the full training set).As ex-pected,models trained on original/new and tested on ori-ginal/new perform better than training and testing on differ-ent annotations.To understand better what the new annota-tions bring to the table,we build a hybrid set of annotations. Pruned annotations is a mid-point that allows to decouple the effects of removing errors and improving alignment. Pruned annotations are generated by matching new and ori-ginal annotations(IoU≥0.5),marking as ignore region any original annotation absent in the new ones,and adding any new annotation absent in the original ones.From original to pruned annotations the main change is re-moving annotation errors,from pruned to new,the main change is better alignment.From table4both ACF and RotatedFilters beneﬁt from removing annotation er-rors,even in MR O−2.This indicates that our new training setFigure 8:Examples of automatically aligned ground truth annotations.Left/right →before/after alignment.1×data 10×data aligned withMR O −2(MR O −4)MR N −2(MR N−4)Orig.Ø19.20(34.28)17.22(31.65)Orig.Orig.10×19.16(32.28)15.94(29.33)Orig.New 1/2×16.97(28.01)14.54(25.06)NewNew 1×16.77(29.76)12.96(22.20)Table 5:Detection quality of RotatedFilters on test set when using different aligned training sets.All mod-els trained with Caltech 10×,composed with different 1×+9×combinations.is better sanitized than the original one.We see in MR N −2that the stronger detector beneﬁts more from better data,and that the largest gain in detection qual-ity comes from removing annotation errors.Alignment beneﬁts The detectors from the ICF family beneﬁt from training with increased training data [16,23],using 10×data is better than 1×(see section 2.1).To lever-age the 9×remaining data using the new 1×annotations we train a model over the new annotations and use this model to re-align the original annotations over the 9×portion.Be-cause the new annotations are better aligned,we expect this model to be able to recover slight position and scale errors in the original annotations.Figure 8shows example results of this process.See supplementary material for details.Table 5reports results using the automatic alignment pro-cess,and a few degraded cases:using the original 10×,self-aligning the original 10×using a model trained over original 10×,and aligning the original 10×using only a fraction of the new annotations (without replacing the 1×portion).The results indicate that using a detector model to improve overall data alignment is indeed effective,and that better aligned training data leads to better detection quality (both in MR O and MR N ).This is in line with the analysis of section 3.2.Already using a model trained on 1/2of the new annotations for alignment,leads to a stronger model than obtained when using original annotations.We name the RotatedFilters model trained using the new annotations and the aligned 9×data,Rotated-Filters-New10×.This model also reaches high me-dian true positives IoU in table 3,indicating that indeed it obtains more precise detections at test time.Conclusion Using high quality annotations for training improves the overall detection quality,thanks both to im-proved alignment and to reduced annotation errors.4.2.Convnets for pedestrian detectionThe results of section 3.2indicate that there is room for improvement by focusing on the core background versus foreground discrimination task (the “classiﬁcation part of object detection”).Recent work [15,20]showed compet-itive performance with convolutional neural networks (con-vnets)for pedestrian detection.We include convnets into our analysis,and explore to what extent performance is driven by the quality of the detection proposals.AlexNet and VGG We consider two convnets.1)The AlexNet from [15],and 2)The VGG16model from [12].Both are pre-trained on ImageNet and ﬁne-tuned over Cal-tech 10×(original annotations)using SquaresChnFtrs proposals.Both networks are based on open source,and both are instances of the R-CNN framework [13].Albeit their training/test time architectures are slightly different (R-CNN versus Fast R-CNN),we expect the result differ-ences to be dominated by their respective discriminative power (VGG16improves 8pp in mAP over AlexNet in the Pascal detection task [13]).Table 6shows that as we improve the quality of the detection proposals,AlexNet fails to provide a consistent gain,eventually worsening the results of our ICF detect-ors (similar observation done in [15]).Similarly VGG provides large gains for weaker proposals,but as the pro-posals improve,the gain from the convnet re-scoring even-tually stalls.After closer inspection of the resulting curves (see sup-plementary material),we notice that both AlexNet and VGG push background instances to lower scores,and at the same time generate a large number of high scoring false positives.The ICF detectors are able to provide high recall proposals,where false positives around the objects have low scores (see [15,supp.material,ﬁg.9]),however convnets have difﬁculties giving low scores to these windows sur-rounding the true positives.In other words,despite their ﬁne-tuning,the convnet score maps are “blurrier”than the proposal ones.We hypothesise this is an intrinsic limita-tion of the AlexNet and VGG architectures,due to their in-ternal feature pooling.Obtaining “peakier”responses from a convnet most likely will require using rather different ar-chitectures,possibly more similar to the ones used for se-mantic labelling or boundaries estimation tasks,which re-quire pixel-accurate output.Fortunately,we can compensate for the lack of spatial resolution in the convnet scoring by using bounding box regression.Adding bounding regression over VGG,and ap-plying a second round of non-maximum suppression (ﬁrst NMS on the proposals,second on the regressed boxes),has。

用于疾病和病症分析的无细胞DNA甲基化模式[发明专利]

专利名称：用于疾病和病症分析的无细胞DNA甲基化模式专利类型：发明专利
发明人：向红·婕思敏·周,康舒里,李文渊,史蒂文·杜比尼特,李青娇
申请号：CN201780047763.3
申请日：20170607
公开号：CN110168099A
公开日：
20190823
专利内容由知识产权出版社提供
摘要：本文公开了利用测序读取来检测并定量由血液样品制备的无细胞DNA中组织类型或癌症类型的存在的方法和系统。

申请人：加利福尼亚大学董事会,南加利福尼亚大学
地址：美国加利福尼亚州
国籍：US
代理机构：北京柏杉松知识产权代理事务所(普通合伙)
更多信息请下载全文后查看。

基于边界约束粒子滤波的多UUV纯方位协同目标跟踪

基于边界约束粒子滤波的多UUV纯方位协同目标跟踪
韩博;徐红丽;邱少雄;张文睿;茹敬雨
【期刊名称】《水下无人系统学报》
【年(卷),期】2024(32)2
【摘要】面向海上跨域协同中多无人水下航行器(UUV)协同探测水面目标需求,针对现有纯方位目标跟踪算法所面临的滤波器初始化困难和水声数据传输丢包问题,提出了一种基于边界约束粒子滤波的多UUV协同纯方位目标跟踪算法。

首先提出了主从式协同探测模型,利用跟随者向领航者上报状态估计结果进行数据融合。

其次,基于UUV传感器和目标的先验信息设计了初始阶段可靠粒子生成方法和更新阶段的指标函数粒子权重优化方法。

最后提出了基于灰色预测的分布式融合算法,得到目标预测结果。

仿真实验将所提算法和其他常见算法进行对比,在通信丢包以及噪声干扰情况下验证了算法的有效性和可行性。

【总页数】10页(P250-259)
【作者】韩博;徐红丽;邱少雄;张文睿;茹敬雨
【作者单位】东北大学信息科学与工程学院
【正文语种】中文
【中图分类】TJ634;U674.941
【相关文献】
1.改进高斯混合粒子滤波的纯方位目标跟踪算法
2.基于改进粒子滤波算法的纯方位目标跟踪
3.邻域迭代重采样粒子滤波的纯方位目标跟踪
4.新型粒子滤波算法及其在纯方位目标跟踪中的应用
5.基于改进遗传粒子滤波的纯方位机动目标跟踪
因版权原因，仅展示原文概要，查看原文内容请购买。

phd多目标跟踪原理

phd多目标跟踪原理
PHD（概率假设密度）多目标跟踪是一种有效的多目标跟踪方法，其基本原理是通过概率假设密度滤波对多目标进行跟踪。

PHD滤波器利用一阶统计矩近似方法对多目标后验概率密度函数求集合积分运算，得到多目标强度即PHD。

PHD方法计算的是多目标联合分布的一阶矩，即将集合积分运算简化为单个变量的积分运算，因此具备了实际运算的可行性。

由于PHD方法属于随机集框架的范畴，所以它也具备随机集方法的优点，即可以避免数据关联过程。

在多目标跟踪过程中，PHD滤波器避免了数据关联的问题，在去除杂波的同时能够实现对目标的联合检测和跟踪。

它适用于关联过程比较复杂的非传统意义下的多目标追踪问题，比如群目标追踪，在密集目标或者杂波背景下对所感兴趣的目标进行检测和跟踪等。

PHD在给定状态空间区域S上的积分是区域S内目标个数的期望值，而PHD的峰值点所对应的状态点可认为是目标状态的估计值。

近些年来，对PHD多目标跟踪算法的研究逐步成为该领域中的一个热点。

本论文针对基于粒子滤波的PHD多目标跟踪方法进行研究，以包含多个目标的模拟仿真和真实红外图像作为主要研究对象，介绍了粒子滤波算法和概
率假设密度滤波算法的理论基础，将这两种方法结合起来应用于红外序列图像的多目标跟踪问题，并对方法的应用进行了实验仿真和分析。

以上内容仅供参考，如需了解PHD多目标跟踪原理的更多信息，建议查阅机器视觉和人工智能相关学术文献或研究资料。

近藤效应

Tunable Kondo eﬀect in a single donor atomnsbergen 1,G.C.Tettamanzi 1,J.Verduijn 1,N.Collaert 2,S.Biesemans 2,M.Blaauboer 1,and S.Rogge 11Kavli Institute of Nanoscience,Delft University of Technology,Lorentzweg 1,2628CJ Delft,The Netherlands and2InterUniversity Microelectronics Center (IMEC),Kapeldreef 75,3001Leuven,Belgium(Dated:September 30,2009)The Kondo eﬀect has been observed in a single gate-tunable atom.The measurement device consists of a single As dopant incorporated in a Silicon nanostructure.The atomic orbitals of the dopant are tunable by the gate electric ﬁeld.When they are tuned such that the ground state of the atomic system becomes a (nearly)degenerate superposition of two of the Silicon valleys,an exotic and hitherto unobserved valley Kondo eﬀect appears.Together with the “regular”spin Kondo,the tunable valley Kondo eﬀect allows for reversible electrical control over the symmetry of the Kondo ground state from an SU(2)-to an SU(4)-conﬁguration.The addition of magnetic impurities to a metal leads to an anomalous increase of their resistance at low tem-perature.Although discovered in the 1930’s,it took until the 1960’s before this observation was satisfactorily ex-plained in the context of exchange interaction between the localized spin of the magnetic impurity and the de-localized conduction electrons in the metal [1].This so-called Kondo eﬀect is now one of the most widely stud-ied phenomena in condensed-matter physics [2]and plays a mayor role in the ﬁeld of nanotechnology.Kondo ef-fects on single atoms have ﬁrst been observed by STM-spectroscopy and were later discovered in a variety of mesoscopic devices ranging from quantum dots and car-bon nanotubes to single molecules [3].Kondo eﬀects,however,do not only arise from local-ized spins:in principle,the role of the electron spin can be replaced by another degree of freedom,for example or-bital momentum [4].The simultaneous presence of both a spin-and an orbital degeneracy gives rise to an exotic SU(4)-Kondo eﬀect,where ”SU(4)”refers to the sym-metry of the corresponding Kondo ground state [5,6].SU(4)Kondo eﬀects have received quite a lot of theoret-ical attention [6,7],but so far little experimental work exists [8].The atomic orbitals of a gated donor in Si consist of linear combinations of the sixfold degenerate valleys of the Si conduction band.The orbital-(or more speciﬁ-cally valley)-degeneracy of the atomic ground state is tunable by the gate electric ﬁeld.The valley splitting ranges from ∼1meV at high ﬁelds (where the electron is pulled towards the gate interface)to being equal to the donors valley-orbit splitting (∼10-20meV)at low ﬁelds [9,10].This tunability essentially originates from a gate-induced quantum conﬁnement transition [10],namely from Coulombic conﬁnement at the donor site to 2D-conﬁnement at the gate interface.In this article we study Kondo eﬀects on a novel exper-imental system,a single donor atom in a Silicon nano-MOSFET.The charge state of this single dopant can be tuned by the gate electrode such that a single electron (spin)is localized on the pared to quantum dots (or artiﬁcial atoms)in Silicon [11,12,13],gated dopants have a large charging energy compared to the level spac-ing due to their typically much smaller size.As a result,the orbital degree of freedom of the atom starts to play an important role in the Kondo interaction.As we will argue in this article,at high gate ﬁeld,where a (near)de-generacy is created,the valley index forms a good quan-tum number and Valley Kondo [14]eﬀects,which have not been observed before,appear.Moreover,the Valley Kondo resonance in a gated donor can be switched on and oﬀby the gate electrode,which provides for an electri-cally controllable quantum phase transition [15]between the regular SU(2)spin-and the SU(4)-Kondo ground states.In our experiment we use wrap-around gate (FinFET)devices,see Fig.1(a),with a single Arsenic donor in the channel dominating the sub-threshold transport charac-teristics [16].Several recent experiments have shown that the ﬁngerprint of a single dopant can be identiﬁed in low-temperature transport through small CMOS devices [16,17,18].We perform transport spectroscopy (at 4K)on a large ensemble of FinFET devices and select the few that show this ﬁngerprint,which essentially consists of a pair of characteristic transport resonances associ-ated with the one-electron (D 0)-and two-electron (D −)-charge states of the single donor [16].From previous research we know that the valley splitting in our Fin-FET devices is typically on the order of a few meV’s.In this Report,we present several such devices that are in addition characterized by strong tunnel coupling to the source/drain contacts which allows for suﬃcient ex-change processes between the metallic contacts and the atom to observe Kondo eﬀects.Fig.1b shows a zero bias diﬀerential conductance (dI SD /dV SD )trace at 4.2K as a function of gate volt-age (V G )of one of the strongly coupled FinFETs (J17).At the V G such that a donor level in the barrier is aligned with the Fermi energy in the source-drain con-tacts (E F ),electrons can tunnel via the level from source to drain (and vice versa)and we observe an increase in the dI SD /dV SD .The conductance peaks indicated bya r X i v :0909.5602v 1 [c o n d -m a t .m e s -h a l l ] 30 S e p 2009FIG.1:Coulomb blocked transport through a single donor in FinFET devices(a)Colored Scanning Electron Micrograph of a typical FinFET device.(b)Diﬀerential conductance (dI SD/dV SD)versus gate voltage at V SD=0.(D0)and(D−) indicate respectively the transport resonances of the one-and two-electron state of a single As donor located in the Fin-FET channel.Inset:Band diagram of the FinFET along the x-axis,with the(D0)charge state on resonance.(c)and(d) Colormap of the diﬀerential conductance(dI SD/dV SD)as a function of V SD and V G of samples J17and H64.The red dots indicate the(D0)resonances and data were taken at1.6 K.All the features inside the Coulomb diamonds are due to second-order chargeﬂuctuations(see text).(D0)and(D−)are the transport resonances via the one-electron and two-electron charge states respectively.At high gate voltages(V G>450mV),the conduction band in the channel is pushed below E F and the FET channel starts to open.The D−resonance has a peculiar double peak shape which we attribute to capacitive coupling of the D−state to surrounding As atoms[19].The current between the D0and the D−charge state is suppressed by Coulomb blockade.The dI SD/dV SD around the(D0)and(D−)resonances of sample J17and sample H64are depicted in Fig.1c and Fig.1d respectively.The red dots indicate the po-sitions of the(D0)resonance and the solid black lines crossing the red dots mark the outline of its conducting region.Sample J17shows aﬁrst excited state at inside the conducting region(+/-2mV),indicated by a solid black line,associated with the valley splitting(∆=2 mV)of the ground state[10].The black dashed lines indicate V SD=0.Inside the Coulomb diamond there is one electron localized on the single As donor and all the observable transport in this regionﬁnds its origin in second-order exchange processes,i.e.transport via a vir-tual state of the As atom.Sample J17exhibits three clear resonances(indicated by the dashed and dashed-dotted black lines)starting from the(D0)conducting region and running through the Coulomb diamond at-2,0and2mV. The-2mV and2mV resonances are due to a second or-der transition where an electron from the source enters one valley state,an the donor-bound electron leaves from another valley state(see Fig.2(b)).The zero bias reso-nance,however,is typically associated with spin Kondo eﬀects,which happen within the same valley state.In sample H64,the pattern of the resonances looks much more complicated.We observe a resonance around0mV and(interrupted)resonances that shift in V SD as a func-tion of V G,indicating a gradual change of the internal level spectrum as a function of V G.We see a large in-crease in conductance where one of the resonances crosses V SD=0(at V G∼445mV,indicated by the red dashed elipsoid).Here the ground state has a full valley degen-eracy,as we will show in theﬁnal paragraph.There is a similar feature in sample J17at V G∼414mV in Fig.1c (see also the red cross in Fig.1b),although that is prob-ably related to a nearby defect.Because of the relative simplicity of its diﬀerential conductance pattern,we will mainly use data obtained from sample J17.In order to investigate the behavior at the degeneracy point of two valley states we use sample H64.In the following paragraphs we investigate the second-order transport in more detail,in particular its temper-ature dependence,ﬁne-structure,magneticﬁeld depen-dence and dependence on∆.We start by analyzing the temperature(T)dependence of sample J17.Fig.2a shows dI SD/dV SD as a function of V SD inside the Coulomb diamond(at V G=395mV) for a range of temperatures.As can be readily observed from Fig.2a,both the zero bias resonance and the two resonances at V SD=+/-∆mV are suppressed with increasing T.The inset of Fig.2a shows the maxima (dI/dV)MAX of the-2mV and0mV resonances as a function of T.We observe a logarithmic dependence on T(a hallmark sign of Kondo correlations)at both resonances,as indicated by the red line.To investigate this point further we analyze another sample(H67)which has sharper resonances and of which more temperature-dependent data were obtained,see Fig.2c.This sample also exhibits the three resonances,now at∼-1,0and +1mV,and the same strong suppression by tempera-ture.A linear background was removed for clarity.We extracted the(dI/dV)MAX of all three resonances forFIG.2:Electrical transport through a single donor atom in the Coulomb blocked region(a)Diﬀerential conductance of sample J17as a function of V SD in the Kondo regime(at V G=395mV).For clarity,the temperature traces have been oﬀset by50nS with respect to each other.Both the resonances with-and without valley-stateﬂip scale similarly with increasing temperature. Inset:Conductance maxima of the resonances at V SD=-2mV and0mV as a function of temperature.(b)Schematic depiction of three(out of several)second-order processes underlying the zero bias and±∆resonances.(c)Diﬀerential conductance of sample H67as a function of V SD in the Kondo regime between0.3K and6K.A linear(and temperature independent) background on the order of1µS was removed and the traces have been oﬀset by90nS with respect to each other for clarity.(d)The conductance maxima of the three resonances of(c)normalized to their0.3K value.The red line is aﬁt of the data by Eq.1.all temperatures and normalized them to their respective(dI/dV)MAX at300mK.The result is plotted in Fig.2d.We again observe that all three peaks have the same(log-arithmic)dependence on temperature.This dependenceis described well by the following phenomenological rela-tionship[20](dI SD/dV SD)max (T)=(dI SD/dV SD)T 2KT2+TKs+g0(1)where TK =T K/√21/s−1,(dI SD/dV SD)is the zero-temperature conductance,s is a constant equal to0.22 [21]and g0is a constant.Here T K is the Kondo tem-perature.The red curve in Fig.2d is aﬁt of Eq.(1)to the data.We readily observe that the dataﬁt well and extract a T K of2.7K.The temperature scaling demon-strates that both the no valley-stateﬂip resonance at zero bias voltage and the valley-stateﬂip-resonance atﬁnite bias are due to Kondo-type processes.Although a few examples ofﬁnite-bias Kondo have been reported[15,22,23],the corresponding resonances (such as our±∆resonances)are typically associated with in-elastic cotunneling.Aﬁnite bias between the leads breaks the coherence due to dissipative transitions in which electrons are transmitted from the high-potential-lead to the low-potential lead[24].These dissipative4transitions limit the lifetime of the Kondo-type processes and,if strong enough,would only allow for in-elastic events.In the supporting online text we estimate the Kondo lifetime in our system and show it is large enough to sustain theﬁnite-bias Kondo eﬀects.The Kondo nature of the+/-∆mV resonances points strongly towards a Valley Kondo eﬀect[14],where co-herent(second-order)exchange between the delocalized electrons in the contacts and the localized electron on the dopant forms a many-body singlet state that screens the valley index.Together with the more familiar spin Kondo eﬀect,where a many-body state screens the spin index, this leads to an SU(4)-Kondo eﬀect,where the spin and charge degree of freedom are fully entangled[8].The ob-served scaling of the+/-∆-and zero bias-resonances in our samples by a single T K is an indication that such a fourfold degenerate SU(4)-Kondo ground state has been formed.To investigate the Kondo nature of the transport fur-ther,we analyze the substructure of the resonances of sample J17,see Fig.2a.The central resonance and the V SD=-2mV each consist of three separate peaks.A sim-ilar substructure can be observed in sample H67,albeit less clear(see Fig.2c).The substructure can be explained in the context of SU(4)-Kondo in combination with a small diﬀerence between the coupling of the ground state (ΓGS)-and theﬁrst excited state(ΓE1)-to the leads.It has been theoretically predicted that even a small asym-metry(ϕ≡ΓE1/ΓGS∼=1)splits the Valley Kondo den-sity of states into an SU(2)-and an SU(4)-part[25].Thiswill cause both the valley-stateﬂip-and the no valley-stateﬂip resonances to split in three,where the middle peak is the SU(2)-part and the side-peaks are the SU(4)-parts.A more detailed description of the substructure can be found in the supporting online text.The split-ting between middle and side-peaks should be roughly on the order of T K[25].The measured splitting between the SU(2)-and SU(4)-parts equals about0.5meV for sample J17and0.25meV for sample H67,which thus corresponds to T K∼=6K and T K∼=3K respectively,for the latter in line with the Kondo temperature obtained from the temperature dependence.We further note that dI SD/dV SD is smaller than what we would expect for the Kondo conductance at T<T K.However,the only other study of the Kondo eﬀect in Silicon where T K could be determined showed a similar magnitude of the Kondo signal[12].The presence of this substructure in both the valley-stateﬂip-,and the no valley-stateﬂip-Kondo resonance thus also points at a Valley Kondo eﬀect.As a third step,we turn our attention to the magnetic ﬁeld(B)dependence of the resonances.Fig.3shows a colormap plot of dI SD/dV SD for samples J17and H64 both as a function of V SD and B at300mK.The traces were again taken within the Coulomb diamond.Atﬁnite magneticﬁeld,the central Kondo resonances of both de-vices split in two with a splitting of2.2-2.4mV at B=FIG.3:Colormap plot of the conductance as a function of V SD and B of sample J17at V G=395mV(a)and H64at V G=464mV(b).The central Kondo resonances split in two lines which are separated by2g∗µB B.The resonances with a valley-stateﬂip do not seem to split in magneticﬁeld,a feature we associate with the diﬀerent decay-time of parallel and anti-parallel spin-conﬁgurations of the doubly-occupied virtual state(see text).10T.From theoretical considerations we expect the cen-tral Valley Kondo resonance to split in two by∆B= 2g∗µB B if there is no mixing of valley index(this typical 2g∗µB B-splitting of the resonances is one of the hall-marks of the Kondo eﬀect[24]),and to split in three (each separated by g∗µB B)if there is a certain degree of valley index mixing[14].Here,g∗is the g-factor(1.998 for As in Si)andµB is the Bohr magneton.In the case of full mixing of valley index,the valley Kondo eﬀect is expected to vanish and only spin Kondo will remain [25].By comparing our measured magneticﬁeld splitting (∆B)with2g∗µB B,weﬁnd a g-factor between2.1and 2.4for all three devices.This is comparable to the result of Klein et al.who found a g-factor for electrons in SiGe quantum dots in the Kondo regime of around2.2-2.3[13]. The magneticﬁeld dependence of the central resonance5indicates that there is no signiﬁcant mixing of valley in-dex.This is an important observation as the occurrence of Valley Kondo in Si depends on the absence of mix-ing(and thus the valley index being a good quantum number in the process).The conservation of valley in-dex can be attributed to the symmetry of our system. The large2D-conﬁnement provided by the electricﬁeld gives strong reason to believe that the ground-andﬁrst excited-states,E GS and E1,consist of(linear combi-nations of)the k=(0,0,±kz)valleys(with z in the electricﬁeld direction)[10,26].As momentum perpen-dicular to the tunneling direction(k x,see Fig.1)is con-served,also valley index is conserved in tunneling[27]. The k=(0,0,±k z)-nature of E GS and E1should be as-sociated with the absence of signiﬁcant exchange interac-tion between the two states which puts them in the non-interacting limit,and thus not in the correlated Heitler-London limit where singlets and triplets are formed.We further observe that the Valley Kondo resonances with a valley-stateﬂip do not split in magneticﬁeld,see Fig.3.This behavior is seen in both samples,as indicated by the black straight solid lines,and is most easily ob-served in sample J17.These valley-stateﬂip resonances are associated with diﬀerent processes based on their evo-lution with magneticﬁeld.The processes which involve both a valleyﬂip and a spinﬂip are expected to shift to energies±∆±g∗µB B,while those without a spin-ﬂip stay at energies±∆[14,25].We only seem to observe the resonances at±∆,i.e.the valley-stateﬂip resonances without spinﬂip.In Ref[8],the processes with both an orbital and a spinﬂip also could not be observed.The authors attribute this to the broadening of the orbital-ﬂip resonances.Here,we attribute the absence of the processes with spinﬂip to the diﬀerence in life-time be-tween the virtual valley state where two spins in seperate valleys are parallel(τ↑↑)and the virtual state where two spins in seperate valleys are anti-parallel(τ↑↓).In con-trast to the latter,in the parallel spin conﬁguration the electron occupying the valley state with energy E1,can-not decay to the other valley state at E GS due to Pauli spin blockade.It wouldﬁrst needs toﬂip its spin[28].We have estimatedτ↑↑andτ↑↓in our system(see supporting online text)andﬁnd thatτ↑↑>>h/k b T K>τ↑↓,where h/k b T K is the characteristic time-scale of the Kondo pro-cesses.Thus,the antiparallel spin conﬁguration will have relaxed before it has a change to build up a Kondo res-onance.Based on these lifetimes,we do not expect to observe the Kondo resonances associated with both an valley-state-and a spin-ﬂip.Finally,we investigate the degeneracy point of valley states in the Coulomb diamond of sample H64.This degeneracy point is indicated in Fig.1d by the red dashed ellipsoid.By means of the gate electrode,we can tune our system onto-or oﬀthis degeneracy point.The gate-tunability in this sample is created by a reconﬁguration of the level spectrum between the D0and D−-charge states,FIG.4:Colormap plot of I SD at V SD=0as a function of V G and B.For increasing B,a conductance peak develops around V G∼450mV at the valley degeneracy point(∆= 0),indicated by the dashed black line.Inset:Magneticﬁeld dependence of the valley degeneracy point.The resonance is ﬁxed at zero bias and its magnitude does not depend on the magneticﬁeld.probably due to Coulomb interactions in the D−-states. Figure4shows a colormap plot of I SD at V SD=0as a function of V G and B(at0.3K).Note that we are thus looking at the current associated with the central Kondo resonance.At B=0,we observe an increasing I SD for higher V G as the atom’s D−-level is pushed toward E F. As B is increased,the central Kondo resonance splits and moves away from V SD=0,see Fig.3.This leads to a general decrease in I SD.However,at around V G= 450mV a peak in I SD develops,indicated by the dashed black line.The applied B-ﬁeld splits oﬀthe resonances with spin-ﬂip,but it is the valley Kondo resonance here that stays at zero bias voltage giving rise to the local current peak.The inset of Fig.4shows the single Kondo resonance in dI SD/dV SD as a function of V SD and B.We observe that the magnitude of the resonance does not decrease signiﬁcantly with magneticﬁeld in contrast to the situation at∆=0(Fig.3b).This insensitivity of the Kondo eﬀect to magneticﬁeld which occurs only at∆= 0indicates the profound role of valley Kondo processes in our structure.It is noteworthy to mention that at this speciﬁc combination of V SD and V G the device can potentially work as a spin-ﬁlter[6].We acknowledge fruitful discussions with Yu.V. Nazarov,R.Joynt and S.Shiau.This project is sup-ported by the Dutch Foundation for Fundamental Re-search on Matter(FOM).6[1]Kondo,J.,Resistance Minimum in Dilute Magnetic Al-loys,Prog.Theor.Phys.3237-49(1964)[2]Hewson,A.C.,The Kondo Problem to Heavy Fermions(Cambridge Univ.Press,Cambridge,1993).[3]Wingreen N.S.,The Kondo eﬀect in novel systems,Mat.Science Eng.B842225(2001)and references therein.[4]Cox,D.L.,Zawadowski,A.,Exotic Kondo eﬀects in met-als:magnetic ions in a crystalline electricﬁeld and tun-neling centers,Adv.Phys.47,599-942(1998)[5]Inoshita,T.,Shimizu, A.,Kuramoto,Y.,Sakaki,H.,Correlated electron transport through a quantum dot: the multiple-level eﬀect.Phys.Rev.B48,14725-14728 (1993)[6]Borda,L.Zar´a nd,G.,Hofstetter,W.,Halperin,B.I.andvon Delft,J.,SU(4)Fermi Liquid State and Spin Filter-ing in a Double Quantum Dot System,Phys.Rev.Lett.90,026602(2003)[7]Zar´a nd,G.,Orbitalﬂuctuations and strong correlationsin quantum dots,Philosophical Magazine,86,2043-2072 (2006)[8]Jarillo-Herrero,P.,Kong,J.,van der Zant H.S.J.,Dekker,C.,Kouwenhoven,L.P.,De Franceschi,S.,Or-bital Kondo eﬀect in carbon nanotubes,Nature434,484 (2005)[9]Martins,A.S.,Capaz,R.B.and Koiller,B.,Electric-ﬁeldcontrol and adiabatic evolution of shallow donor impuri-ties in silicon,Phys.Rev.B69,085320(2004)[10]Lansbergen,G.P.et al.,Gate induced quantum conﬁne-ment transition of a single dopant atom in a Si FinFET, Nature Physics4,656(2008)[11]Rokhinson,L.P.,Guo,L.J.,Chou,S.Y.,Tsui, D.C.,Kondo-like zero-bias anomaly in electronic transport through an ultrasmall Si quantum dot,Phys.Rev.B60, R16319-R16321(1999)[12]Specht,M.,Sanquer,M.,Deleonibus,S.,Gullegan G.,Signature of Kondo eﬀect in silicon quantum dots,Eur.Phys.J.B26,503-508(2002)[13]Klein,L.J.,Savage, D.E.,Eriksson,M.A.,Coulombblockade and Kondo eﬀect in a few-electron silicon/silicon-germanium quantum dot,Appl.Phys.Lett.90,033103(2007)[14]Shiau,S.,Chutia,S.and Joynt,R.,Valley Kondo eﬀectin silicon quantum dots,Phys.Rev.B75,195345(2007) [15]Roch,N.,Florens,S.,Bouchiat,V.,Wernsdirfer,W.,Balestro, F.,Quantum phase transistion in a single molecule quantum dot,Nature453,633(2008)[16]Sellier,H.et al.,Transport Spectroscopy of a SingleDopant in a Gated Silicon Nanowire,Phys.Rev.Lett.97,206805(2006)[17]Calvet,L.E.,Wheeler,R.G.and Reed,M.A.,Observa-tion of the Linear Stark Eﬀect in a Single Acceptor in Si, Phys.Rev.Lett.98,096805(2007)[18]Hofheinz,M.et al.,Individual charge traps in siliconnanowires,Eur.Phys.J.B54,299307(2006)[19]Pierre,M.,Hofheinz,M.,Jehl,X.,Sanquer,M.,Molas,G.,Vinet,M.,Deleonibus S.,Oﬀset charges acting as ex-cited states in quantum dots spectroscopy,Eur.Phys.J.B70,475-481(2009)[20]Goldhaber-Gordon,D.,Gres,J.,Kastner,M.A.,Shtrik-man,H.,Mahalu, D.,Meirav,U.,From the Kondo Regime to the Mixed-Valence Regime in a Single-Electron Transistor,Phys.Rev.Lett.81,5225(1998) [21]Although the value of s=0.22stems from SU(2)spinKondo processes,it is valid for SU(4)-Kondo systems as well[8,25].[22]Paaske,J.,Rosch,A.,W¨o lﬂe,P.,Mason,N.,Marcus,C.M.,Nyg˙ard,Non-equilibrium singlet-triplet Kondo ef-fect in carbon nanotubes,Nature Physics2,460(2006) [23]Osorio, E.A.et al.,Electronic Excitations of a SingleMolecule Contacted in a Three-Terminal Conﬁguration, Nanoletters7,3336-3342(2007)[24]Meir,Y.,Wingreen,N.S.,Lee,P.A.,Low-TemperatureTransport Through a Quantum Dot:The Anderson Model Out of Equilibrium,Phys.Rev.Lett.70,2601 (1993)[25]Lim,J.S.,Choi,M-S,Choi,M.Y.,L´o pez,R.,Aguado,R.,Kondo eﬀects in carbon nanotubes:From SU(4)to SU(2)symmetry,Phys.Rev.B74,205119(2006) [26]Hada,Y.,Eto,M.,Electronic states in silicon quan-tum dots:Multivalley artiﬁcial atoms,Phys.Rev.B68, 155322(2003)[27]Eto,M.,Hada,Y.,Kondo Eﬀect in Silicon QuantumDots with Valley Degeneracy,AIP Conf.Proc.850,1382-1383(2006)[28]A comparable process in the direct transport throughSi/SiGe double dots(Lifetime Enhanced Transport)has been recently proposed[29].[29]Shaji,N.et.al.,Spin blockade and lifetime-enhancedtransport in a few-electron Si/SiGe double quantum dot, Nature Physics4,540(2008)7Supporting InformationFinFET DevicesThe FinFETs used in this study consist of a silicon nanowire connected to large contacts etched in a60nm layer of p-type Silicon On Insulator.The wire is covered with a nitrided oxide(1.4nm equivalent SiO2thickness) and a narrow poly-crystalline silicon wire is deposited perpendicularly on top to form a gate on three faces.Ion implantation over the entire surface forms n-type degen-erate source,drain,and gate electrodes while the channel protected by the gate remains p-type,see Fig.1a of the main article.The conventional operation of this n-p-n ﬁeld eﬀect transistor is to apply a positive gate voltage to create an inversion in the channel and allow a current toﬂow.Unintentionally,there are As donors present be-low the Si/SiO2interface that show up in the transport characteristics[1].Relation between∆and T KThe information obtained on T K in the main article allows us to investigate the relation between the splitting (∆)of the ground(E GS)-andﬁrst excited(E1)-state and T K.It is expected that T K decreases as∆increases, since a high∆freezes out valley-stateﬂuctuations.The relationship between T K of an SU(4)system and∆was calculated by Eto[2]in a poor mans scaling approach ask B T K(∆) B K =k B T K(∆=0)ϕ(2)whereϕ=ΓE1/ΓGS,withΓE1andΓGS the lifetimes of E1and E GS respectively.Due to the small∆com-pared to the barrier height between the atom and the source/drain contact,we expectϕ∼1.Together with ∆=1meV and T K∼2.7K(for sample H67)and∆=2meV and T K∼6K(for sample J17),Eq.2yields k B T K(∆)/k B T K(∆=0)=0.4and k B T K(∆)/k B T K(∆= 0)=0.3respectively.We can thus conclude that the rela-tively high∆,which separates E GS and E1well in energy, will certainly quench valley-stateﬂuctuations to a certain degree but is not expected to reduce T K to a level that Valley eﬀects become obscured.Valley Kondo density of statesHere,we explain in some more detail the relation be-tween the density of states induced by the Kondo eﬀects and the resulting current.The Kondo density of states (DOS)has three main peaks,see Fig.1a.A central peak at E F=0due to processes without valley-stateﬂip and two peaks at E F=±∆due to processes with valley-state ﬂip,as explained in the main text.Even a small asym-metry(ϕclose to1)will split the Valley Kondo DOS into an SU(2)-and an SU(4)-part[3],indicated in Fig1b in black and red respectively.The SU(2)-part is positioned at E F=0or E F=±∆,while the SU(4)-part will be shifted to slightly higher positive energy(on the order of T K).A voltage bias applied between the source and FIG.1:(a)dI SD/dV SD as a function of V SD in the Kondo regime(at395mV G)of sample J17.The substructure in the Kondo resonances is the result of a small diﬀerence between ΓE1andΓGS.This splits the peaks into a(central)SU(2)-part (black arrows)and two SU(4)-peaks(red arrows).(b)Density of states in the channel as a result ofϕ(=ΓE1/ΓGS)<1and applied V SD.drain leads results in the Kondo peaks to split,leaving a copy of the original structure in the DOS now at the E F of each lead,which is schematically indicated in Fig.1b by a separate DOS associated with each contact.The current density depends directly on the density of states present within the bias window deﬁned by source/drain (indicated by the gray area in Fig1b)[4].The splitting between SU(2)-and SU(4)-processes will thus lead to a three-peak structure as a function of V SD.Figure.1a has a few more noteworthy features.The zero-bias resonance is not positioned exactly at V SD=0, as can also be observed in the transport data(Fig1c of the main article)where it is a few hundredµeV above the Fermi energy near the D0charge state and a few hundredµeV below the Fermi energy near the D−charge state.This feature is also known to arise in the Kondo strong coupling limit[5,6].We further observe that the resonances at V SD=+/-2mV diﬀer substantially in magnitude.This asymmetry between the two side-peaks can actually be expected from SU(4)Kondo sys-tems where∆is of the same order as(but of course al-ways smaller than)the energy spacing between E GS and。

基于深度学习注意力机制的肝脏肿瘤图像分割方法[发明专利]

专利名称：基于深度学习注意力机制的肝脏肿瘤图像分割方法专利类型：发明专利
发明人：李春国,陆敬奔,张翅,冷天然,高振,孙希茜,杨绿溪
申请号：CN202111359719.0
申请日：20211117
公开号：CN114677403A
公开日：
20220628
专利内容由知识产权出版社提供
摘要：本发明公开了一种基于深度学习注意力机制的肝脏肿瘤图像分割方法，属于图像处理领域。

本发明将轴向注意力和多尺度注意力融合进深度学习肝脏肿瘤分割网络。

轴向注意力能够在占用少量计算资源的情况下，有效地提取肝脏肿瘤CT图像的全局信息特征，而多尺度注意力可以有效地针对多尺度目标进行自适应的特征提取。

网络整体采用U型的网络结构，主干为卷积提取路径，辅路为注意力机制，能够有效地提高肝脏肿瘤分割的性能。

申请人：东南大学
地址：211102 江苏省南京市江宁区东南大学路2号
国籍：CN
代理机构：南京瑞弘专利商标事务所(普通合伙)
代理人：秦秋星
更多信息请下载全文后查看。

改进的多目标GM-PHD分量融合算法

收稿日期：2020-01-05修回日期：2020-02-07基金项目：河南省科技攻关资助项目（182102210116）作者简介：孙志强（1983-），男，河南安阳人，讲师。

研究方向：自动化控制技术。

*摘要：密集杂波的平行多目标跟踪场景中，高斯混合概率假设密度滤波器的计算代价随着分量的增多而不断变大，且其目标状态估计精度较低。

为了解决这些问题，基于高斯混合概率假设密度滤波框架，提出一种改进的目标分量融合算法。

通过目标分量的权重、均值及协方差的充分协作，该算法能够极大程度地融合目标强度中的相似分量，同时能够有效地避免真实目标分量被错误融合。

仿真结果表明，密集杂波环境下该算法不仅具有较高的目标状态估计精度，而且其计算代价相对较低。

关键词：目标跟踪，高斯混合概率假设密度，分量融合，运算代价中图分类号：TN953文献标识码：ADOI ：10.3969/j.issn.1002-0640.2021.02.019引用格式：孙志强.改进的多目标GM-PHD 分量融合算法［J ］.火力与指挥控制，2021，46（2）：109-113.改进的多目标GM-PHD 分量融合算法*孙志强（商丘职业技术学院机电系，河南商丘476000）Improved Component Merging Algorithm for Multi-target GM-PHDSUN Zhi-qiang（Department of Mechanical and Electronic Engineering ，Shangqiu Polytechnic ，Shangqiu 476000，China ）Abstract ：In parallel multiple target tracking scenes with dense clutters ，the computation cost ofthe Gaussian Mixture Probability Hypothesis Density （GM -PHD ）filter varies with the increase of components ，and its target state estimate accuracy is low.To overcome these problems ，an improved target component merging algorithm under the framework of the GM-PHD filter is presented.By the full collaboration of the weight ，mean value and covariance of target component ，the proposed algorithm can greatly merge the similar components in the target intensity ，and effectively avoid incorrectly fusion of components of real targets.The simulation results show that the proposed algorithm not only has highaccuracy of target state estimation but also has a relatively low computation cost in dense clutter environments.Key words ：target tracking ，gaussian mixture probability hypothesis density ，component merging ，computation costCitation format ：SUN Z Q.Improved component merging algorithm for multi-target GM-PHD ［J ］.Fire Control &Command Control ，2021，46（2）：109-113.0引言近年来，基于随机有限集（Random Finite Set ，RFS ）的概率假设密度（Probability Hypothesis Density ，PHD ）［1］滤波器引起了目标跟踪领域众多学者的密切关注。

入侵检测技术

攻击都来自内部，对于企业内部心怀不满旳员工来说，防火墙形同虚设； – 不能提供实时入侵检测能力，而这一点，对于目前层出不穷旳攻击技术来说是至关主要旳； – 对于病毒等束手无策。
IDS旳功能与作用
• 辨认黑客常用入侵与攻击手段。入侵检测系统经过分析多种攻击特征，能够全方面迅速地辨认探测攻击、拒绝服务攻击、缓冲区溢出攻击、电子邮件攻击、浏览器攻击等多种常用攻击手段，并做相应旳防范和向管理员发出警告
内容
• 入侵检测技术旳概念 • 入侵检测系统旳功能 • 入侵检测技术旳分类 • 入侵检测技术旳原理、构造和流程 • 入侵检测技术旳将来发展
基本概念
• 入侵检测技术是为确保计算机系统旳安全而设计与配置旳一种能够及时发觉并报告系统中未授权或异常现象旳技术，是一种用于检测计算机网络中违反安全策略行为旳技术。
• 监控网络异常通信。 IDS系统会对网络中不正常旳通信连接做出反应，确保网络通信旳正当性；任何不符合网络安全策略旳网络数据都会被 IDS侦测到并警告。
IDS旳功能与作用
• 鉴别对系统漏洞及后门旳利用。 • 完善网络安全管理。 IDS经过对攻击或入侵旳
检测及反应，能够有效地发觉和预防大部分旳网络入侵或攻击行为，给网络安全管理提供了一种集中、以便、有效旳工具。使用IDS系统旳监测、统计分析、报表功能，能够进一步完善网管。
• 1996年， GRIDS(Graph-based Intrusion Detection System)设计和实现处理了入侵检测系统伸缩性不足旳问题，使得对大规模自动或协同攻击旳检测更为便利。 Forrest 等人将免疫原理用到分布式入侵检测领域
IDS旳发展史
• 1997年， Mark crosbie 和 Gene Spafford将遗传算法利用到入侵检

急性重症肝炎小鼠肝脏趋化因子IP-10 mRNA的表达及意义

ｖｍｉｙ，ｕ５１３ｎ
【ｂｔａｔＯｂｅｔｅＴｎｅｔａｅｔｅｓｎｃｎｅｏｅａｉｃｅｏｉＰ一１ＮｘｒｓｏｉＡｓｒｃ】ｊｃｖｏｉｓｇｔｈｉｉａｃｆｈｐｔｈｍｋｎＩｉｖｉｇｆｉｃｅ０ｍＲＡｅｐｅｓｎｉｍｃｉｎｅ
ｗｔｕｍｉａｔｈｐｔｉ．ＭｅｈｄＭｏｓｄｌｏｌａｔｈｐｔｉｗｓｅｔｂｉｈｄｂｎｒｐｒｔｎａｎｅｔｎｗｔｉｆｌｎｎｅａｉｓｈｔｔｏｓｕｅｍｏｅｆｆｉｎｅａｉｓａｓｌｅｙｉｔｅｌｅｌｉｊｃｉｉｕｍｎｔａｓａｏｏｈ
坏死的重要原因。最近的研究表明，化因子趋
鼠购自湖北省实验动物研究中心。腹腔注射２０ｐ０，Ｌ含１０ＰＵＭＶ一０ＦＨ３的无菌生理盐水，收集感染ＭＨ３Ｖ一Ｏ２、８７、４４、ｈ时的小鼠肝脏组织标本，蜡切片，Ｅ２石Ｈ染色，观察肝组织病理学改变，据Ｋｏｅ方法记算根ｎｄｌ
化因子ＩＰ一１ＲＡ的表达水平。结果０ｍＮ
感染后的Ｂ１／Ｊ小鼠血清ＡＪ的水平以及肝脏ＨＡ积分均显著升ａｂｃＩＴＩ
高，内Ｉ肝Ｐ一１ＲＡ水平也显著上升，４０ｍＮ在８和７２ｈ分别为感染前的９９和１０倍。Ｐａｓｎ相关分析证实感染５ｅｒｏ后肝内趋化因子ＩＰ一１Ｎ的表达与血清ＡＴ及肝组织ＨＡ积分显著正相关（０ｍＲＡＬＩＰ＜００）结论趋化因子．５。

伴侣蛋白10的免疫抑制作用[发明专利]

专利名称：伴侣蛋白10的免疫抑制作用专利类型：发明专利
发明人：G·R·希尔,T·巴诺维奇
申请号：CN200380108296.9
申请日：20031106
公开号：CN1735428A
公开日：
20060215
专利内容由知识产权出版社提供
摘要：本发明是针对cpn10在移植中的应用，特别是针对在治疗和/防止移植物抗宿主疾病中的应用。

本发明提供了将cpn10给予供体和/或受体动物或细胞、来自供体的组织或器官的方法，尽管本发明特别优选用于治疗供体和受体动物。

该方法进一步包括给予供体和/或受体动物至少一种其他的免疫抑制剂以防止或缓解移植物抗宿主疾病。

申请人：悉生物有限公司
地址：澳大利亚昆士兰
国籍：AU
代理机构：北京纪凯知识产权代理有限公司
更多信息请下载全文后查看。

一种新的多肽-人神经元线蛋白12和编码这种多肽的多核苷酸[发明专利]

专利名称：一种新的多肽－人神经元线蛋白12和编码这种多肽的多核苷酸
专利类型：发明专利
发明人：毛裕民,谢毅
申请号：CN99125727.8
申请日：19991223
公开号：CN1300782A
公开日：
20010627
专利内容由知识产权出版社提供
摘要：本发明公开了一种新的多肽－人神经元线蛋白12,编码此多肽的多核苷酸和经DNA重组技术产生这种多肽的方法。

本发明还公开了此多肽用于治疗多种疾病的方法,如恶性肿瘤,血液病,HIV感染和免疫性疾病和各类炎症等。

本发明还公开了抗此多肽的拮抗剂及其治疗作用。

本发明还公开了编码这种新的入神经元线蛋白12的多核苷酸的用途。

申请人：复旦大学,上海博道基因技术有限公司
地址：200433 上海市邯郸路220路
国籍：CN
代理机构：复旦大学专利事务所
更多信息请下载全文后查看。

粒子群优化人工免疫粒子滤波器

粒子群优化人工免疫粒子滤波器杜正聪;冯大海;牛高远【期刊名称】《四川大学学报（工程科学版）》【年(卷),期】2013(045)001【摘要】为解决粒子滤波算法中存在的粒子退化和样本枯竭问题,提出一种新的粒子滤波算法.利用粒子群优化思想促使采样粒子向高似然区域移动,减缓粒子权值的退化；再通过人工免疫算法中的变异操作扩大算法寻找最优值的范围并增加粒子的多样性,避免算法陷入局部最优,增强算法的全局搜索能力,进而缓解样本枯竭.实验表明,该算法比标准粒子滤波的状态估计精度提高近40倍,比扩展卡尔曼粒子滤波提高近28倍,比无迹卡尔曼粒子滤波提高近6倍,滤波效率为37.523％,是标准粒子滤波的37倍,该算法具有更好的实时性和更高的状态估计精度,能有效缓解粒子的退化和样本的枯竭.【总页数】6页(P146-151)【作者】杜正聪;冯大海;牛高远【作者单位】电子科技大学电子工程学院,四川成都610054;攀枝花学院,四川攀枝花617000;攀枝花学院,四川攀枝花617000;西华大学电气信息学院,四川成都610039;攀枝花学院,四川攀枝花617000;西华大学电气信息学院,四川成都610039【正文语种】中文【中图分类】TN957.51【相关文献】1.基于Metropolis-Hastings变异的粒子群优化粒子滤波器 [J], 路威;张邦宁2.限制速度粒子群优化和自适应速度粒子群优化在无约束优化问题中的应用 [J], 许君;鲁海燕;石桂娟3.人工免疫系统及人工免疫遗传算法在优化中的应用 [J], 郑德玲;梁瑞鑫;付冬梅;李晓刚;方彤4.基于人工免疫粒子群优化算法的动态聚类分析 [J], 王磊;吉欢;徐庆征5.基于Rao-Blackwellized粒子滤波器移动机器人SLAM研究 [J], 黄辉;邹安安;胡鹏;邹媛媛;蔡庆荣因版权原因，仅展示原文概要，查看原文内容请购买。

改进的粒子滤波人体目标跟踪算法

改进的粒子滤波人体目标跟踪算法
徐胜;黄晁;孙松
【期刊名称】《无线电通信技术》
【年(卷),期】2018(44)1
【摘要】针对现有基于粒子滤波(PF)的行人目标跟踪算法跟踪精度不高、速度慢以及遮挡鲁棒性不好的问题,提出一种结合支持向量机(SVM)检测的改进跟踪算法.在跟踪的置信度小于阈值时进行行人跟踪目标的再检测,以校正跟踪位置.对粒子群优化(PSO)算法在优化时可能陷入局部解的现状,进行混沌粒子优化(CPSO)寻优全局解.实验结果表明,提出的算法在一定的粒子数目前提下精度优于其他基于粒子滤波的行人目标跟踪算法,有效降低PF所需粒子数,算法可实时跟踪.
【总页数】4页(P69-72)
【作者】徐胜;黄晁;孙松
【作者单位】宁波大学信息科学与工程学院,浙江宁波 315211;宁波中国科学院信息技术应用研究院,浙江宁波 315040;宁波中科集成电路设计中心有限公司,浙江宁波 315040;宁波大学信息科学与工程学院,浙江宁波 315211
【正文语种】中文
【中图分类】TN911.73
【相关文献】
1.基于粒子滤波的视频目标跟踪算法研究及改进 [J], 毛玮;韩旭;夏志强
2.基于改进粒子滤波的稀疏子空间单目标跟踪算法 [J], 宫海洋;任红格;史涛;李福
进
3.改进的粒子滤波目标跟踪算法 [J], 高海;韩洋
4.结合匈牙利指派和改进粒子滤波的多目标跟踪算法 [J], 李华楠;曹林;王东峰;付冲
5.基于特征匹配与改进粒子滤波的冠脉目标跟踪算法 [J], 王光磊;卢倩;刘秀玲;王鹏宇
因版权原因，仅展示原文概要，查看原文内容请购买。

基于融合反馈式粒子滤波器的多目标跟踪算法

基于融合反馈式粒子滤波器的多目标跟踪算法
付钿; 牛玉刚
【期刊名称】《《华东理工大学学报（自然科学版）》》
【年(卷),期】2010(036)002
【摘要】针对足球视频中多球员跟踪问题,提出一种基于融合反馈式粒子滤波的多目标跟踪算法。

该算法将球员状态的检测信息与球员的动力学模型结合,设计粒子滤波算法的建议分布,使其能融合最新的观测数据,并由此对下一帧球员可能的状态进行抽样,然后计算各抽样与参考模板间的相似度,最终通过估计球员状态来达到跟踪目的。

仿真实验结果表明:本文提出的跟踪算法能较好地解决不同球员间的遮挡问题,实现多名球员的跟踪,具有较好的实时性和鲁棒性。

【总页数】6页(P261-266)
【作者】付钿; 牛玉刚
【作者单位】华东理工大学信息科学与工程学院上海 200237
【正文语种】中文
【中图分类】TP391
【相关文献】
1.基于UKF的Rao-Blackwellized粒子滤波器多传感器多目标跟踪算法研究 [J], 袁志勇;顾晓东
2.一种基于交互式粒子滤波器的视频中多目标跟踪算法 [J], 刘晨光;程丹松;刘家锋;黄剑华;唐降龙
3.基于相机雷达融合的改进GM-PHD多目标跟踪算法 [J], 张晗; 李森; 白傑
4.基于特征融合的复杂场景多目标跟踪算法研究 [J], 王志余
5.基于多特征融合的多目标跟踪算法研究 [J], 王帅;苍岩
因版权原因，仅展示原文概要，查看原文内容请购买。

基于粒子群算法优化的音频特征应用研究

基于粒子群算法优化的音频特征应用研究
王志强;郭宁;傅向华
【期刊名称】《计算机科学》
【年(卷),期】2014(041)010
【摘要】在深入研究音频特征的基础上,提取响度特征和音调特征,并利用粒子群算法优化特征权重.提出一种对歌唱片段进行自动评价的方法,用于视频点歌系统的实时评分模块.实验结果表明,该系统能够反映演唱者歌声和歌曲原唱两者内容的相似程度,从而给出了有效的评分标准.
【总页数】5页(P45-49)
【作者】王志强;郭宁;傅向华
【作者单位】深圳大学计算机与软件学院深圳518060;深圳大学计算机与软件学院深圳518060;深圳大学计算机与软件学院深圳518060
【正文语种】中文
【中图分类】TP301
【相关文献】
1.量子粒子群算法优化最小二乘支持向量机及其应用研究 [J], 刘倩
2.改进粒子群算法优化支持向量机在故障诊断中的应用研究 [J], 孙瑶琴
3.基于细菌觅食特征改进粒子群算法优化SVM模型参数研究 [J], 李宝晨;金赛赛;仝蕊;连光耀
4.基于粒子群算法优化LVQ神经网络的应用研究 [J], 张超;魏三强;胡秀建;梁西陈
5.改进的粒子群算法优化的特征选择方法 [J], 李炜;巢秀琴
因版权原因，仅展示原文概要，查看原文内容请购买。

关注社交异配性的社交机器人检测框架

关注社交异配性的社交机器人检测框架
余尚戎;肖景博;殷琪林;卢伟
【期刊名称】《信息网络安全》
【年(卷),期】2024()2
【摘要】随着社交机器人的迭代,其倾向于与正常用户进行更多交互,对其检测变得更具挑战性。

现有检测方法大多基于同配性假设,由于忽视了不同类用户间存在的联系,难以保持良好的检测性能。

针对这一问题文章提出一种关注社交异配性的社交机器人检测框架,以社交网络用户间的联系为依据,通过充分挖掘用户社交信息来应对异配影响,并实现更精准的检测。

文章分别在同配视角和异配视角下看待用户之间的联系,将社交网络构建为图,通过消息传递机制实现同配边和异配边聚合,以提取节点的频率特征,同时利用图中各节点特征聚合得到社交环境特征,将以上特征混合后用于检测。

实验结果表明,文章所提方法在开源数据集上的检测效果优于基线方法,证明了该方法的有效性。

【总页数】9页(P319-327)
【作者】余尚戎;肖景博;殷琪林;卢伟
【作者单位】中山大学计算机学院;中山大学信息技术教育部重点实验室;广东省信息安全技术重点实验室
【正文语种】中文
【中图分类】TP309
【相关文献】
1.“类人”社交机器人检测数据集扩充方法研究
2.社交机器人参与的在线社交网络竞争性信息传播模型及仿真
3.高中化学校本作业研发策略分析
4.基于多维动态特征验证的社交机器人账号检测
5.结合主动学习与关系图卷积神经网络的社交机器人检测
因版权原因，仅展示原文概要，查看原文内容请购买。

1、下载文档前请自行甄别文档内容的完整性，平台不提供额外的编辑、内容补充、找答案等附加服务。
2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
3、如文档侵犯您的权益，请联系客服反馈,我们会尽快为您处理(人工客服工作时间：9:00-18:30)。

Particle Filter
MVUD
Search for new objects
Frame # t-1 Resample Frame #t DIP Learning Re-weight
Output
MVUD
Infer
DIP
Particle
Video Source
DCP Learning
Occlusion Model Response
...
LNl
DCP
Tree structured multi-view human upper body detector.
A. Multi-View Upper-Body Detector Part detectors have been proved to be an effective way to detect and track partially occluded objects [10]. Generally speaking, the smaller a part is, the larger probability it will be fully visible. From this prospective, a smaller object part has a larger traceability. However, smaller object part detector becomes harder to learn since it provides less information for learning. Although employing multiple part detectors of different size will remedy this problem [10], the computation cost will increase simultaneously. In this paper, we only train one part detector covering the upper-body area which is the most informative region of the human body. To deal with the view variances of the upper-body, the training samples are divided into three different views, i.e., frontalrear, left proﬁle and right proﬁle, and trained using the method in [3]. The multi-view upper-body detector provides a very discriminative model. Figure 2 shows the structure of the detector. For details about the training process, please see [3]. B. Online Discriminative Learning Although part detector could detect many partially occluded humans, it is likely to fail when more serious occlusions happen which prevent the part region from fully visible. What is more, the general human detector tends to drift when humans are close to each other due to its congenital deﬁciency at distinguishing different humans. To address these problems, an online learning process is proposed to effective collect the discriminative features of each human and be used to track a human under more serious dense situations. During the online discriminative learning process, two different types of features are explored, the discriminative interest points and the discriminative color patch. The interest points are those have an expressive texture in their respective localities which provide the local information of one object and could be visible in very dense situation, while the color patch could be one salient image region (e.g. the clothes region) which provides the global information of one object and could be used to re-track the object after long time full occlusion.
Multiple Human Tracking Based on Multi-View Upper-Body Detection and Discriminative Learning
Junliang Xing, Haizhou Ai Department of Computer Science and Technology Tsinghua University Beijing 100084, China Email: ahz@ Shihong Lao Core Technology Center Omron Corporation Kyoto 619-0283, Japan Email: lao@ari.ncl.omron.co.jp
Abstract—This paper focuses on the problem of tracking multiple humans in dense environments which is very challenging due to recurring occlusions between different humans. To cope with the difﬁculties it presents, an ofﬂine boosted multiview upper-body detector is used to automatically initialize a new human trajectory and is capable of dealing with partial human occlusions. What is more, an online learning process is proposed to learn discriminative human observations, including discriminative interest points and color patches, to effectively track each human when even more occlusions occur. The ofﬂine and online observation models are neatly integrated into the particle ﬁlter framework to robustly track multiple highly interactive humans. Experiments results on CAVIAR dataset as well as many other challenging real-world cases demonstrate the effectiveness of the proposed method. Keywords-object tracking; object detection; discriminative learning; particle ﬁlter;
Discriminative Learning
DCP
Tracker Output
Figure 1.
System overview.
I. I NTRODUCTION Multiple object tracking in video is of fundamental importance for many applications, such as visual surveillance, trafﬁc safety monitoring, human computer interaction, etc. This could be an easy task when the objects are isolated from each other in a relatively clean background. However, real-world cases often go against this assumption by posing a complex background and serious occlusions among different objects. To track multiple objects in complex situations, some early methods track motion blobs and regard each individual blob as one human [5], [11]. These methods usually assume the background is ﬁxed and use background subtraction [7] to provide relatively robust object motion blobs. The foreground blob based methods are not discriminative and is likely to fail when the background changes suddenly. Recently, object detection researches have resulted in many promising detectors of particular object classes, e.g., faces [9] and humans or pedestrians [2], [10]. They can provide good observations for detection-based tracking algorithms. By applying object detectors into particle ﬁlter [4] framework, impressive results of tracking one single object have been achieved in [6]. In multiple object tracking, detection based methods suffer from the occlusion problem which