| 摘 要: 为实现当归叶片病虫害检测需求,提出了一种基于改进Res2Net模型的当归叶片病虫害图像识别算法。首先,在骨干网络中引入CBAM(Convolutional Block Attention Module)卷积注意力机制;其次,调整残差块结构,优化数据归一化分布;最后,优化分类器结构,在全局平均池化层后增加全连接层,提高模型的表达能力。经验证:改进模型的准确率为98.52%,较原Res2Net模型提升1.98个百分点;精确率、召回率和F1值分别提升了2.09%、1.94%和2.01%。研究结果表明,该模型为当归叶片病虫害的精准防控和智能管理提供了技术支撑。 |
| 关键词: 当归病虫害 Res2Net 特征提取 注意力机制 图像识别 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 甘肃省重点研发计划项目(21YF5GA088) |
|
| Research on Image Classification of Angelica Leaf Diseases and Pests Based on Improved Res2Net Model |
|
KONG Yali,LI Jiazhen,WANG Lianguo
|
(College of Information and Technology, Gansu Agricultural University, Lanzhou 730070, China)
2802448740@qq.com; 1553705843@qq.com; wanglg@gsau.edu.cn
|
| Abstract: To meet the demand for detecting diseases and pests in angelica leaves, this study proposes an image recognition algorithm for angelica leaf diseases and pests based on an improved Res2Net model. First, the Convolutional Block Attention Module (CBAM) was introduced into the backbone network. Second, the residual block structure was adjusted to optimize the data normalization distribution. Finally, the classifier structure was optimized by adding a fully connected layer after the global average pooling layer to enhance the model’s expressive ability.Experimental results show that the accuracy of the improved model reaches 98.52% , which is 1.98 percentage points higher than the original Res2Net model. The precision, recall, and F1-score increase by 2.09% , 1.94% , and 2.01% ,respectively. The findings indicate that this model provides technical support for the precise prevention and intelligent management of angelica leaf diseases and pests. |
| Keywords: angelica diseases and pests Res2Net feature extraction attention mechanism image recognition |
