基于共享卷积与多尺度增强注意力的学生课堂行为检测算法

打开文本图片集

中图分类号：TP391.4；G434 文献标识码：A 文章编号：1006-8228（2026）01-52-06

Abstract：AimingatthelimitationsoftheYOLOllmodelinstudentclasroombehaviordetection，thispaperproposesan improvedYOLOlln-SGMalgorithmbasedonsharedconvolutionandmultiscaleenhancedatention.Toaddresstheinsufficient featureextractionofsmalltargetsinthebackrowofclasroomscenes，aSharedConvolutionFeaturePyramid（SCFP）moduleis designedinthebackbonenetworkofYOLO11ByreplacingordinaryconvolutionsinYOLOllwithGolden CudgelConvolution （GCBlock），thedetectionauracyisimprovedandthenumberofparametersiscontroledAdditionally，Multi-ScaleEanced AtentionDetectionHead（MSDH）isdesignedtoenhancethemodel'sabilitytoextractfeaturesofocudedtargets.Experimental resultsshowthatcomparedwiththe baseline modelYOLOlln，the proposedYOLOi1n-SGMmodelachievesimprovementsof 2.2% in precision， 1.8% in recall， 2.0% in mAP50，and 3.1% in mAP50-95.It significantly enhances detection accuracy while effectively controlling the number ofparameters.

Keywords：YOLOl1；Classroom BehaviorDetection;Shared Convolution；Multi-ScaleEnhanced Attention

0引言

2025年4月由教育部等九部门联合印发《关于加快推进教育数字化的意见》，明确提出“全面推进智能化，促进人工智能助力教育变革”。（剩余7525字）