融合多尺度注意力机制的棉花枯萎病识别算法研究  

Research on Cotton Blight Recognition Algorithm Based on Multi-scale Attention Mechanism

作　　者：李文雪 孟洪兵[1,2] 孙丽丽 韩璐宇 LI Wenxue;MENG Hongbing;SUN Lili;HAN Luyu(College of Information Engineering,Tarim University,Alaer 843300,China;Tarim Key Laboratory of Oasis Agriculture,Ministry of Education(Tarim University),Alaer 843300,China)

机构地区：[1]塔里木大学信息工程学院,新疆阿拉尔843300 [2]塔里木绿洲农业教育部重点实验室(塔里木大学),新疆阿拉尔843300

出　　处：《现代农业装备》2025年第2期86-92,共7页Modern Agricultural Equipment

基　　金：新疆生产建设兵团财政科技计划项目(1121DB008)。

摘　　要：针对棉花枯萎病检测时叶片背景复杂、有遮挡以及病变多尺度等问题,提出一种融合多尺度注意力机制的YOLOv7棉花枯萎病识别算法。为适应棉花枯萎病病斑尺度不一、形状多变的特征,提升识别检测效果,在YOLOv7的特征提取网络中添加多尺度注意力模块,通过多尺度信息的融合和自适应权重调整机制,提高模型的泛化性能;同时为了降低模型的计算量和参数量,提高模型的运行速度,更换特征提取网络为InceptionNeXt。试验结果表明,改进后的YOLOv7模型检测精度P达到95.9%,对比基线模型提升了2.3%;平均精度mAP@0.5达到88.15%,提高了3.67%;召回率R达到94.65%,提升了2.31%;参数量为33.73 M,减少了2.78 M;计算量为89.65 G,降低了14.62 G;证明该算法能有效提高棉花枯萎病的识别精度和效率,可对棉花病害防治提供一定的技术支撑。Aiming at the problems such as complex leaf background,occlusion and multi-scale lesions in cotton blight detection,a novel YOLOv7 cotton blight recognition algorithm combining multi-scale attention mechanism was proposed.In order to adapt to the different scales and shapes of cotton blight spots and improve the recognition and detection effect,a multi-scale attention module was first added to the feature extraction network of YOLOv7,and the generalization performance of the model was improved through multi-scale information fusion and adaptive weight adjustment mechanism.At the same time,in order to reduce the calculation amount and parameter number of the model and improve the running speed of the model,the feature extraction network was changed to InceptionNeXt.Experimental results showed that the detection accuracy of the improved YOLOv7 model was up to 95.9%,2.3%higher than that of the baseline model,and the average accuracy mAP@0.5 was up to 88.15%,3.67%higher.The recall rate R reached 94.65%,got an increase of 2.31%.The number of parameters decreased by 2.78 M to 33.73 M,and the calculation amount decreased by 14.62 G to 89.65 G.The results showed that the improved algorithm could effectively improve the accuracy and efficiency of cotton blight identification,and provided some technical support for disease control.

关 键 词：YOLOv7 棉花 注意力机制 枯萎病 病害检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]