高光谱图像弱小目标深度定位算法

打开文本图片集

引用格式：，.高光谱图像弱小目标深度定位算法[J].现代电子技术，2025，48（18）：29-33.

关键词：高光谱图像；弱小目标；深度定位；光谱匹配；自适应阈值；YOLOv5中图分类号：TN911.73-34；TP751.1 文献标识码：A 文章编号：1004-373X（2025）18-0029-05

Abstract：Inorder toavoid theconcealmentof dimandsmalltarget featuresbyhyperspectral noiseandcomplex backgrounds，reducedataredundancyinthelocalization processandenhance thecomprehensiveness ofweakandsmlltarget localization，adeeplocalizationalgorithmfordimandsmalltargetsinhyperspectralimagesisproposed.Basedonspectral matchingmethod，thehyperspectralimagesareobtainedandtheirdimensionalitiesarereduced.Theadaptivethresholdalgorithm isusedtopreprocessthehyperspectralimageafterdimensionalityreduction，anddetectdimandsmalltargetstoobtainan enhanced hyperspectral imageofdimnandsmalltargets.BasedontheYOLOv5neural network algorithm，depth localizationof dimandsmalltargetsinenhancedhyperspectral images.Theexperimentalresultsshowthattheproposedalgorithmcangenerate clearerhyperspectraltargetimagesbasedonhyperspectralsequencedata，effectivelyreducetheirdimensionalityanddetect multipledimandsmalltargets，obtainingaccuratelocalizationofalldimandsmalltargets.TheKappacoeficientsof the localization resultsare all greater than O.87，proving thatthe algorithm has high reliabilityand comprehensiveness.

KeyWords：hyperspectral image;dimandsmalltarget;deeplocalization;spectral matching;adaptivethreshold;YOLO

0 引言

高光谱图像弱小自标深度定位将图像光谱信息与空间信息相融合，通过光谱差异和空间上下文关联，实现复杂背景中小目标的像素级定位；但由于高光谱图像是三维数据，其中涵盖上百个连续波段，容易导致目标定位效率下降；且在低光照等条件下，目标光谱信号容易被背景噪声淹没，导致原始光谱信噪比降低。（剩余5021字）