引用格式:黄玉玲，陶昕辰，朱涛，等．残差对抗目标检测算法的遥感图像检测［J］．电光与控制，2023，30(7):63-67．HUANGYL,TAOXC,ZHUT,etal．Aremotesensingimagedetectionmethodbasedonresidualsadversarialobjectdetectionalgorithm［J］．ElectronicsOptics＆Control,2023,30(7):63-67．残差对抗目标检测算法的遥感图像检测黄玉玲1，陶昕辰1，朱涛1，司俊文1，吕昌东1，吴迪1，沈占锋1，2(1．苏州大学，江苏苏州215000;2．中国科学院空天信息创新研究院，北京100000)摘要:针对遥感图像目标检测的目标尺度小、分辨率过低的问题，提出残差对抗目标检测算法的遥感图像检测方法。通过残差对抗的方式对图像的特征信息进行重构，完成图像分辨率的提升。在Backbone主干网络提取图像特征信息的基础上由Neck结构将特征进行融合，最后由CIoU_Loss损失函数提高定位回归精度，提高模型性能。实验结果表明，与其他算法相比，在平均精确率、平均召回率、平均综合指标F1值、平均mAP值方面分别提高了8．15%，6．9%，7．15%，6．75%。所提算法在低分辨率遥感图像目标检测方面准确性较高，对遥感图像小目标检测效果较好。关键词:遥感;目标检测;超分辨率;残差对抗中图分类号:P237文献标志码:Adoi:10．3969/j．issn．1671－637X．2023．07．011AＲemoteSensingImageDetectionMethodBasedonＲesidualsAdversarialObjectDetectionAlgorithmHUANGYuling1,TAOXinchen1,ZHUTao1,SIJunwen1,LYUChangdong1,WUDi1,SHENZhanfeng1,2(1．SoochowUniversity,Suzhou215000,China;2．AerospaceInformationＲesearchInstitute,ChineseAcademyofSciences,Beijing100000,China)Abstract:Aimingattheproblemsofsmallobjectscaleandlowresolutioninremotesensingimageobjectdetection,aremotesensingimagedetectionmethodbasedonresidualadversarialobjectdetectionalgorithmisproposedinthispaper．Thefeatureinformationoftheimageisreconstructedthroughresidualadversarialism,soastorealizetheimprovementofimageresolution．BasedonimagefeatureinformationwhichisextractedbyBackbonenetwork,thefeaturesarefusedbyNeckstructure．Finally,theCIoU_Lossfunctionisdesignedtoincreasetheregressionaccuracy,andimprovemodelperformance．Experimentalresultsshowthat,comparedwithotheralgorithms,themeanprecision,meanrecall,meanF1-scoreandmeanmAPvalueofthisalgorithmareimpr...
