| 基于WGAN和MCA-MobileNet的番茄叶片病害识别 |
| |
| 引用本文： | 王志强,于雪莹,杨晓婧,兰玉彬,金鑫宁,马景余.基于WGAN和MCA-MobileNet的番茄叶片病害识别[J].农业机械学报,2023,54(5):244-252. |
| |
| 作者姓名： | 王志强 于雪莹 杨晓婧 兰玉彬 金鑫宁 马景余 |
| |
| 作者单位： | 山东理工大学 |
| |
| 基金项目： | 山东省引进顶尖人才“一事一议”专项经费项目（鲁政办字[2018]27号） |
| |
| 摘 要： | 针对番茄病害识别模型参数量大、计算成本高、准确率低等问题，本文提出一种基于多尺度特征融合和坐标注意力机制的轻量级网络（Multi scale feature fusion and coordinate attention MobileNet, MCA-MobileNet）模型。采集10类番茄叶片图像，采用基于Wasserstein距离的生成对抗网络（Wasserstein generative adversarial networks, WGAN）进行数据增强，解决了样本数据不足和不均衡的问题，提高模型的泛化能力。在原始模型MobileNet-V2的基础上，引入改进后的多尺度特征融合模块对不同尺度的特征图进行特征提取，提高模型对不同尺度的适应性；将轻量型的坐标注意力机制模块（Coordinate attention, CA）嵌入倒置残差结构中，使模型更加关注叶片中的病害特征，提高对病害种类的识别准确率。试验结果表明，MCA-MobileNet对番茄叶片病害的识别准确率达到94.11%，较原始模型提高2.84个百分点，且参数量仅为原始模型的1/6。该方法较好地平衡了模型的识别准确率和计算成本，为番茄叶片病害的现场部署和实时检测提供了思路和技术支撑。
|
| 关 键 词： | 番茄叶片病害 深度学习 生成对抗网络 多尺度特征融合 注意力机制 |
| 收稿时间： | 2022/10/10 0:00:00 |
Tomato Leaf Diseases Recognition Based on WGAN and MCA-MobileNet |
| |
| WANG Zhiqiang,YU Xueying,YANG Xiaojing,LAN Yubin,JIN Xinning,MA Jingyu.Tomato Leaf Diseases Recognition Based on WGAN and MCA-MobileNet[J].Transactions of the Chinese Society of Agricultural Machinery,2023,54(5):244-252. |
| |
| Authors: | WANG Zhiqiang YU Xueying YANG Xiaojing LAN Yubin JIN Xinning MA Jingyu |
| |
| Institution: | Shandong University of Technology |
| |
| Abstract: | With the continuous development of smart agricultural technology, plant disease identification models are increasingly pursuing the goals of accuracy, efficiency, and light weight. Aiming at the problems of large number of parameters, high calculation cost and low accuracy in the current tomato disease recognition model, a lightweight network was proposed based on multi-scale feature fusion and coordinate attention mechanism (Multi-scale feature fusion and coordinate attention MobileNet, MCA-MobileNet) model. Totally ten types of tomato leaf images were collected, and Wasserstein generative adversarial networks (WGAN) based on Wasserstein distance for data enhancement was used, which solved the problem of insufficient and unbalanced sample data and improved the generalization ability of the model. On the basis of the original model MobileNet-V2, an improved multi-scale feature fusion module was introduced to extract features from feature maps of different scales to improve the adaptability of the model to different scales;the lightweight coordinate attention mechanism module (Coordinate attention, CA) embedded in the inverted residual structure, so that the model paid more attention to the disease characteristics in the leaves and improved the recognition accuracy of the disease types. The test results showed that the accuracy rate of MCA-MobileNet for identifying tomato leaf diseases reached 94.11%, which was 2.84 percentage points higher than that of the original model, and the number of parameters was only 1/6 of the original model. This method better balanced the recognition accuracy and calculation cost of the model, and provided ideas and technical support for field deployment and real-time detection of tomato leaf diseases. |
| |
| Keywords: | tomato leaf disease deep learning WGAN multi-scale feature fusion coordinate attention |
|
| 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 |
|
