温室蓝莓花期蜜蜂授粉行为的轻量化视觉检测方法研究

打开文本图片集

中图分类号：TP391；S663.9 文献标识码：A 文章编号：2095-5553（2026）02-0268-08

Abstract：Aiming atthe precisedetectionand recognitionof polinationbehaviorof blueberry bees during flowering period in greenhouse，a lightweight smal-target visual detection method basedon YOLO—POL was proposed．The EfcientViT network with improvedSGAMatentionreconstructionoptimizationwasused toreplace the backbone featureextraction network，greatlyreducing thenumberof parametersin themodel，andrealizing the layer-by-layer extractionofkeyfeaturesof beepollination images.The lightweightmodule RepghostC3wasfurtherdesignedto improve the neck network，maintaintheability toextractsmall targetfeature informationof thenetwork，andeffectivelyreduce the network computational complexity and parameter number.The predictive frame generation mechanism of Anchor Free was introduced toreduce thenumberof multi-scale detectors to reduce the computational overhead．Theresults showed that the mAP of the proposed method was 96.2% ，the precision was 97.1% ，and the number of network parameters was reduced by 31.94% compared with YOLOv5s. Compared with other lightweight detection networks，it achievesacomprehensive balanceof detectionaccuracy，model parametersand robustness，which is more suitable for deploymentinactual agriculturalapplicationdetectionscenarios，andhasstrong practicalvalueforfutureapplicationsof mobile terminals.

Keywords：greenhouse blueberies；lightweight；polinationof honeybees；SGAMattention；flowering period

0 引言

温室蓝莓蜜蜂授粉是蓝莓温室种植中至关重要的一个环节。（剩余11956字）