基于改进YOLOv5在电力巡检中的目标检测算法研究游越1，伊力哈木·亚尔买买提1，吕怡凡2，赵子凡3（1.新疆大学电气工程学院，乌鲁木齐830046；2.中央民族大学信息工程学院，北京100081；3.西安理工大学电气工程学院，西安710054）摘要：针对输电线路巡检的复杂场景下被遮挡以及目标较小导致的误检问题，文中提出了一种基于YOLOv5的改进算法模型。首先通过数据增强和数据扩充数据集进行预处理；其次给引入SE模块，加强对不同尺度下目标的特征融合；然后引入CBAM模块，进行进一步特征提取，使提取的特征信息更加突出。最后，对损失函数进行优化，改进小数据导致的样本不均衡现象。经实验证明，改进后的算法有效地提高了被遮挡的目标和小目标的识别率，mAP较原算法精度提高了5%，Recall提高了14.6%。与其他改进模型相比，精度、召回率都有提高，验证了该模型在各种场合下具有较强的鲁棒性和泛化能力。关键词：输电线路巡检；遮挡；YOLOv5；CBAM；SEResearchonTargetDetectionAlgorithmBasedonImprovedYOLOv5inPowerPatrolInspectionYOUYue1，YILIHAMUYaermaimaiti1，LYUYifan2，ZHAOZifan3（1.SchoolofElectricalEngineering，XinjiangUniversity，Urumqi830046，China；2.SchoolofInformationEngineering，MinzuUniversityofChina，Beijing100081，China；3.SchoolofElectricalEngineering，Xi’anUniversityofTechnology，Xi’an710054，China）Abstract:Inviewoffalsedetectionduetoshelteringandsmalltargetincomplexscenesofpatrolinspectionoftrans⁃mission，akindofimprovedalgorithmmodelbasedonYOLOv5isproposedinthispaper.First，preprocessingiscar⁃riedoutthroughdataenhancementanddataexpansiondataset.Then，theSEmoduleisintroducedtostrengthenthefeaturefusionoftargetsatdifferentscales.Afterthat，theCBAMmoduleisintroducedforfeatureextractionfurthersotomakethefeatureinformationmoreprominent.Finally，thelossfunctionisoptimizedtoimprovethesampleim⁃balanceduetosmalldata.Itisprovedthroughexperimentsthattheimprovedalgorithmeffectivelyimprovestherec⁃ognitionrateofshelteringtargetsandsmalltargets.TheaccuracyofmAP，comparedwiththeoriginalalgorithm，isimprovedby5%andRecallisimprovedby14.6%.Comparedwithotherimprovedmodels，bothprecisionandrecallrateareimproved，verifyingthatthemodelhasstr...
