注意机制目标检测算法在核桃果实识别中的应用  

Application of Attention Mechanism Object Detection Algorithm in Walnut Fruit Recognition

作　　者：张弦[1] 周贤惠 李红 蒋伟昌 Zhang Xian;Zhou Xianhui;Li Hong;Jiang Weichang(Ecological Branch of Yunnan Institute of Forest Inventory and Planning,Kunming 650031,Yunnan,China;Dali Branch of Yunnan Institute of Forest Inventory and Planning,Dali 671000,Yunnan,China)

机构地区：[1]云南省林业调查规划院生态分院,云南昆明650031 [2]云南省林业调查规划院大理分院,云南大理671000

出　　处：《绿色科技》2024年第22期237-240,共4页Journal of Green Science and Technology

摘　　要：核桃果实的识别可以应用于大型核桃种植园区的产量预测工作。精准地识别出核桃树上的果实,对提高产量预测准确率具有重要的意义。为了解决目前存在的挑战,提出了一种Attention-YOLO核桃目标检测算法,用于精准识别无人机遥感图像中的核桃,达到为核桃产量预测提供技术支持的目的。首先,引入CBAM注意力强化Backbone网络,提升算法的特征提取能力;然后,使用多重注意力机制模块重塑原始Head层,优化目标定位精度。实验表明:Attention-YOLO算法对核桃的检测精度(mAP0.5)达到了92.6%,召回率(R)达到了96%,相比于其他的主流模型均具有明显优势。The recognition of walnut fruits can be applied to the yield prediction work of large walnut planting parks.Accurately identifying the fruits on walnut trees is of great significance for improving the accuracy of yield prediction.To solve the existing challenges,this paper proposes an Attention-YOLO walnut object detection algorithm for accurately identifying walnuts in unmanned aerial vehicle remote sensing images,so as to achieve the purpose of providing technical support for walnut yield prediction.First,the CBAM attention-enhanced Backbone network is introduced to improve the feature extraction ability of the algorithm.Then,the multiple attention mechanism module is used to reshape the original Head layer and optimize the target positioning accuracy.Experiments show that the detection accuracy(mAP0.5)of the Attention-YOLO algorithm for walnuts reaches 92.6%,and the recall rate(R)reaches 96%.Compared with other mainstream models,it has obvious advantages.

关 键 词：核桃目标检测 产量预测 注意力机制 

分 类 号：S664[农业科学—果树学]