| 郭文娟,冯全.基于改进YOLO的棉花叶片病害检测[J].干旱地区农业研究,2024,(6):195~205 |
| 基于改进YOLO的棉花叶片病害检测 |
| Cotton leaf disease detection based on improved YOLO |
| |
| DOI：10.7606/j.issn.1000-7601.2024.06.20 |
| 中文关键词: 棉花 病害检测 YOLOX-S YOLOv7 注意力机制 |
| 英文关键词:cotton disease detection YOLOX-S YOLOv7 attention mechanism |
| 基金项目:甘肃省教育厅创新基金项目（2022B-144）;甘肃省自然科学基金项目（24JRRA571）;甘肃政法大学校级重点项目（GZF2024XZD07） |
|
| 摘要点击次数: 63 |
| 全文下载次数: 42 |
| 中文摘要: |
| 为实现棉花的智能化管理，针对复杂背景下棉花叶片病害检测准确率低的问题，基于改进YOLO算法对棉花叶片常见病害进行检测。分别在YOLOX-S和YOLOv7算法中引入SE、CBAM和ECA注意力模块，引导检测模型更关注棉花叶片病害区域特征，有效抑制背景噪声干扰，显著降低模型的漏减率。采用梯度加权类激活映射产生目标检测热图，可视化有效特征，理解模型的关注区域。结果表明：YOLOX-S+SE模型、YOLOX-S+CBAM模型和YOLOX-S+ECA模型的检测效果均优于YOLOX-S模型，并且YOLOX-S+CBAM模型的检测性能最优；YOLOv7+SE模型、YOLOv7+CBAM模型和YOLOv7+ECA模型的检测效果均优于YOLOv7模型，其中YOLOv7+CBAM模型的检测效果最好；与其他常见的目标检测模型相比，YOLOX-S+CBAM模型和YOLOv7+CBAM模型的准确率、召回率、综合评价指标、平均准确率值均高于Faster R-CNN模型和SSD模型，综合比较结果，YOLOX-S+CBAM模型的检测性能优于YOLOv7+CBAM模型，实现了检测速度、检测精度和模型体量之间的平衡，能够在复杂背景下对棉花叶片病害有较好的实时检测效果，对棉花病害防治具有重要的指导意义。 |
| 英文摘要: |
| To achieve intelligent management of cotton and address the challenge of low accuracy in detecting cotton leaf diseases in complex environments, this paper presents results from using an improved YOLO algorithm for detecting common cotton leaf diseases. It incorporates the SE, CBAM, and ECA attention modules into the YOLOX-S and YOLOv7 algorithms to enhance the model’s focus on diseased areas, effectively suppress background noise, and significantly reduce the omission rate. Using Gradient Weighted Class Activation Mapping to generate target detection heatmaps, visualize effective features, and understand the region of interest of the model. The experimental results demonstrate that the detection performance of the YOLOX-S+SE, YOLOX-S+CBAM, and YOLOX-S+ECA models surpassed that of the YOLOX-S model, with the YOLOX-S+CBAM model achieving the best performance. Similarly, the YOLOv7+SE, YOLOv7+CBAM, and YOLOv7+ECA models outperformed the YOLOv7 model, with the YOLOv7+CBAM model delivering the best detection performance. Compared to other common object detection models, the YOLOX-S+CBAM and YOLOv7+CBAM models exhibit higher values in accuracy rate, recall rate, comprehensive evaluation index, and average accuracy than the Faster R-CNN and SSD models. A comprehensive comparison shows that the YOLOX-S+CBAM model outperforms the YOLOv7+CBAM model, striking a balance between detection speed, accuracy, and model size. It demonstrates strong real\|time detection capabilities for cotton leaf diseases in complex backgrounds. This research provides valuable guidance for the prevention and control of cotton diseases. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| | |