| 基于StyleGAN2-ADA和改进YOLO v7的葡萄叶片早期病害检测方法 |
| |
| 引用本文： | 张林鍹,巴音塔娜,曾庆松.基于StyleGAN2-ADA和改进YOLO v7的葡萄叶片早期病害检测方法[J].农业机械学报,2024,55(1):241-252. |
| |
| 作者姓名： | 张林鍹 巴音塔娜 曾庆松 |
| |
| 作者单位： | 新疆大学；清华大学 |
| |
| 基金项目： | 新疆维吾尔族自治区自然科学基金项目(2022D01C431) |
| |
| 摘 要： | 为实现葡萄早期病害的快速准确识别，针对葡萄病害的相似表型症状识别率低及小病斑检测困难的问题，以葡萄黑腐病和黑麻疹病为研究对象，提出了一种基于自适应鉴别器增强的样式生成对抗网络与改进的YOLO v7相结合的葡萄黑腐病和黑麻疹病的病斑检测方法。通过自适应鉴别器增强的样式生成对抗网络和拉普拉斯滤波器的方差扩充葡萄病害数据。采用MSRCP算法进行图像增强,改善光照环境凸显病斑特征。以YOLO v7网络框架为基础，将BiFormer注意力机制嵌入特征提取网络，强化目标区域的关键特征；采用BiFPN代替PA-FPN，更好地实现低层细节特征与高层语义信息融合，以同时降低计算复杂度；在YOLO v7的检测头部分嵌入SPD模块，以提高模型对低分辨率图像的检测性能；并采用CIoU与NWD损失函数组合对损失函数重新定义，实现对小目标快速、准确识别。实验结果表明，该方法病斑检测精确率达到94.1%，相比原始算法提升5.7个百分点，与Faster R-CNN、YOLO v3-SPP和YOLO v5x等模型相比分别提高3.3、3.8、4.4个百分点，能够实现葡萄早期病害快速准确识别，对于保障葡萄产业发展具有重要意义。
|
| 关 键 词： | 葡萄 病害识别 StyleGAN2-ADA 目标检测 自注意力机制 YOLO v7 |
| 收稿时间： | 2023/6/1 0:00:00 |
Grape Disease Detection Method Based on StyleGAN2-ADA and Improved YOLO v7 |
| |
| ZHANG Linxuan,BA Yintan,ZENG Qingsong.Grape Disease Detection Method Based on StyleGAN2-ADA and Improved YOLO v7[J].Transactions of the Chinese Society of Agricultural Machinery,2024,55(1):241-252. |
| |
| Authors: | ZHANG Linxuan BA Yintan ZENG Qingsong |
| |
| Abstract: | Black rot and brown spot disease of grapes are diseases that seriously threaten grape yields, and identification of grape diseases early is of great significance for disease prevention and control and grape yield. However, current disease detection methods have a high leakage rate. The black rot and brown spot were taken as the research objects, a method for detecting grape black rot and brown spot based on adaptive discriminator enhanced style generation adversarial network combined with improved YOLO v7 was proposed. Firstly, the grape disease data were expanded by the adaptive discriminator enhanced style generation adversarial network + deblurring processing. Secondly, the MSRCP algorithm was used to enhance the image and improve the lighting environment to highlight the characteristics of disease spots. Finally, based on the YOLO v7 network framework, the BiFormer attention mechanism was embedded in the feature extraction network to strengthen the key features of the target area. BiFPN was used instead of PA-FPN to better realize multi-scale feature fusion and reduce computational complexity. SPD module was introduced in the detection head section of YOLO v7 to improve the detection performance of low-resolution images. The combination of CIoU and NWD loss function was used to redefine the loss function to achieve rapid and accurate identification of small targets. The experimental results showed that the accuracy of spot detection in this method reached 94.1%, which was 5.7 percentage points higher than that of the original algorithm, and 3.3 percentage points, 3.8 percentage points, and 4.4 percentage points higher than that of Faster R-CNN, YOLO v3-SPP, and YOLO v5x models, respectively, which can realize the rapid and accurate identification of early grape diseases, which was of positive significance for ensuring the development of the grape industry. |
| |
| Keywords: | |
|
| 点击此处可从《农业机械学报》浏览原始摘要信息 |
| 点击此处可从《农业机械学报》下载免费的PDF全文 |
|
