文章编号:1002-2082(2024)02-0405-10基于背景抑制和分类校正的小样本目标检测方法蔡伟,王鑫,蒋昕昊,杨志勇,陈栋(火箭军工程大学导弹工程学院,陕西西安710025)摘要:为进一步提升小样本条件下对空中来袭目标的检测识别成功率,提出一种基于背景抑制和分类校正的小样本目标检测方法。首先,针对空中来袭目标背景前景易混淆问题,在区域候选网络前端引入背景抑制模块,通过抑制背景特征和增强前景特征来减轻目标背景对检测的影响;其次,在背景抑制模块后插入特征聚合模块,聚焦目标特征,通过缓解小样本条件导致目标特征难提取、不明显的问题,校正网络模型的分类参数;最后,在检测头网络中引入对比分支,增强了类别内相似性和类别间独特性,缓解来袭目标“类间相似性高,类内差异性大”的问题,实现了对网络分类的进一步校正。实验结果表明,所提出的算法在1、2、3、5、10shot实验中均表现最佳,平均精度分别达到28.3%、32.8%、39.9%、42.9%和56.2%,提升了小样本空中来袭目标的检测性能。关键词:小样本目标检测;空中来袭目标;背景抑制;分类校正;深度学习中图分类号:TN219文献标志码:ADOI:10.5768/JAO202445.0203002FewshottargetdetectionmethodbasedonbackgroundsuppressionandclassificationcorrectionCAIWei,WANGXin,JIANGXinhao,YANGZhiyong,CHENDong(SchoolofMissileEngineering,RocketMilitaryEngineeringUniversity,Xi'an710025,China)Abstract:Tofurtherimprovethesuccessrateofdetectingandidentifyingairbornetargetsunderfewshotconditions,afewshottargetdetectionmethodbasedonbackgroundsuppressionandclassificationcorrectionwasproposed.Firstly,aimingattheproblemthatthebackgroundforegroundofincomingairtargetswaseasytoconfuse,abackgroundsuppressionmodulewasintroducedinthefrontendoftheregionalcandidatenetwork,whichenhancedtheforegroundfeaturesbysuppressingthebackgroundfeaturesandreducedtheinfluenceofthetargetbackgroundondetection.Secondly,thefeatureaggregationmodulewasinsertedafterthebackgroundsuppressionmoduletofocusonthetargetfeatures,andtoalleviatetheproblemthatthetargetfeaturesweredifficulttoextractandnotobviousduetofewshotconditions,soastocorrecttheclassificationparametersofthenetworkmodel.Finally,acontrastbranchwasintroducedintothedetectionh...