基于分段重标定的稠密卷积神经网络的分带染色体图像类型识别-豆柴文库

您所在位置：网站首页 / 基于分段重标定的稠密卷积神经网络的分带染色体图像类型识别.docx / 文档详情

在线预览结束，喜欢就下载吧，查找使用更方便

5 金币

下载文档

/ 2

下载提示

如果您无法下载资料，请参考说明：

1、部分资料下载需要金币，请确保您的账户上有足够的金币

2、已购买过的文档，再次下载不重复扣费

3、资料包下载后请先用软件解压，在使用对应软件打开

文本预览

基于分段重标定的稠密卷积神经网络的分带染色体图像类型识别
Abstract:
Inrecentyears,thedevelopmentofdeeplearninghasgreatlyboostedtheadvancementofimagerecognitiontasks.However,forchromosomeimagetyperecognition,duetotheuniquecharacteristicsofchromosomeimages,existingdeeplearningmethodsoftenencounterchallengesincapturingandextractingsignificantfeatures.Inthispaper,weproposeadenseconvolutionalneuralnetwork(DCNN)basedonpiecewiserecalibrationforchromosomeimagetyperecognition.TheDCNNframeworkconsistsofmultipledenseblocksandtransitionlayerswhichenableeffectivefeatureextractionthroughdenseconnectionsandcompression.Moreover,weintroduceapiecewiserecalibrationschemetoadaptivelyrecalibratetheextractedfeatures,allowingthenetworktofocusonimportantinformationindifferentregionsofthechromosomeimage.Experimentalresultsonachromosomeimagedatasetdemonstratethesuperiorityofourproposedmethodcomparedtoexistingstate-of-the-arttechniquesintermsofrecognitionaccuracyandrobustness.
1.Introduction
Chromosomeimagetyperecognitionplaysacriticalroleinclinicalgeneticsandcytogeneticanalysis.Accuraterecognitionofchromosomeimagetypesisessentialfordiseasediagnosis,geneticcounselingandresearch.Duetotheuniquecharacteristicsofchromosomeimages,suchascomplexstructuresandvariousstainingpatterns,theidentificationofchromosomeimagetypesremainschallenging.
2.RelatedWork
Traditionalmethodsforchromosomeimagetyperecognitionrelyonhandcraftedfeaturesandclassifiers.However,thesemethodssufferfromlimitationsinfeaturerepresentationandlackrobustness.Withtheemergenceofdeeplearning,convolutionalneuralnetworks(CNNs)haveshownpromisingresultsinvariousimagerecognitiontasks.DenseNetisapopularCNNarchitecturethatintroducesdenseconnectionstoenhancefeaturereuseandgradientflow.
3.ProposedMethod
Inthispaper,weproposeadenseconvolutionalneuralnetwork(DCNN)basedonpiecewiserecalibrationforchromosomeimagetyperecognition.TheDCNNconsistsofmultipledenseblocksandtransitionlayers.Eachdenseblockcomprisesmultipledenselyconnectedconvolutionallayers,allowingforeffectivefeatureextra