基于改进的ResNet18模型识别番茄叶片多种病害
Identifying various diseases of tomato leaves based on improved ResNet18 model
杨进进 1张文慧 1王哲1
作者信息
- 1. 华北水利水电大学信息工程学院,郑州 450046
- 折叠
摘要
针对传统番茄叶片病害识别方法效率低、准确性差等问题,提出了一种改进的ResNet18番茄叶片病害识别算法.首先将ResNet18中输入部分的7×7大卷积核替换为3×3小卷积核,减少网络参数数量,增加网络的非线性表达能力.然后在改进的ResNet18网络中加入轻量级卷积块注意力模块(CBAM),增强网络对病害细节特征的提取能力,提高识别精度;并使用单周期余弦退火算法调整学习率,进一步优化网络结构,加快模型收敛效果,提高训练速度.实验以早疫病、棒孢病等9种常见的番茄叶片病害为主要研究对象,在改进模型上的平均识别准确率达到99.60%.结果表明,构建的ResNet18-Ck3x3-CBAM模型可用于番茄叶片病害识别且具有良好的识别效果.
Abstract
In order to solve the problems of low efficiency and poor accuracy of traditional tomato leaf disease identification methods,an improved ResNet18 tomato leaf disease identification algorithm was proposed.First,replace the 7×7 large convolution kernel in the input part of ResNet18 with a 3×3 small convolution kernel to reduce the number of network parameters and increase the nonlinear expression ability of the network.A lightweight convolutional block attention module(CBAM)is added to the ResNet18 network to enhance the network's ability to extract detailed features of the disease and improve the recognition accuracy;and the single-cycle cosine annealing algorithm is used to adjust the learning rate,further optimize the network structure,acceler-ate the model convergence effect,and improve the training speed.The experiment mainly focused on 9 common tomato leaf diseases such as Early_blight and Target_Spot.In terms of recognition accuracy,the average recognition accuracy of the improved model reached 99.60%.The results show that the constructed ResNet18-Ck3×3-CBAM model can be used to identify tomato leaf diseases and has good recognition performance.
关键词
番茄叶片/病害识别/ResNet18/CBAM注意力机制/余弦退火学习率Key words
tomato leaf/disease identification/ResNet18/convolutional block attention module/cosine annealing learing rate引用本文复制引用
出版年
2024
NETL
NSTL
万方数据