基于Cutout-ResNet50的野外环境水稻病害识别系统  

Rice disease identification system in the field environment based on Cutout-ResNet50

作　　者：黄思琪 张正华[1] 郭丽瑞 李斌[1] HUANG Siqi;ZHANG Zhenghua;GUO Lirui;LI Bin(School of Information Engineering(School of Artificial Intelligence),Yangzhou University,Yangzhou 225127,China)

机构地区：[1]扬州大学信息工程学院(人工智能学院),江苏扬州225127

出　　处：《扬州大学学报（自然科学版）》2024年第3期37-45,共9页Journal of Yangzhou University：Natural Science Edition

基　　金：江苏省科技计划(现代农业)重点资助项目(BE2023340);江苏省研究生实践创新计划资助项目(SJCX23_1894);扬州市科技计划(产业前瞻性与关键技术)资助项目(YZ2023004)。

摘　　要：针对水稻病害图像在野外环境下存在的光照不均、明暗变化明显、因遮挡导致目标特征缺失和噪声重叠,以及野外环境的水稻数据集少且质量差等问题,提出一种基于改进ResNet50算法的野外环境水稻病害识别方法,并设计识别系统.在传统ResNet50算法的基础上采用迁移学习技术对学习知识跨领域迁移,缓解数据集样本不足和不均衡造成的过拟合现象;利用Cutout增强方法对特征信息随即筛选,模拟复杂的野外环境,加强算法的泛化能力;对学习率采用余弦退火优化策略,提高算法的稳定性.结果表明:改进的ResNet50算法在小型水稻病害数据集上的识别准确率达97.24%,明显高于传统ResNet50算法,且该改进方法亦能提升VGG16、GoogLeNet和MobileNetV3-large等其他卷积神经网络算法的模型识别性能.将该模型部署于系统,可为水稻病害识别在实际应用工程中的发展提供技术参考.Aiming at the problems of rice disease images in the field environment,such as uneven illumination,obvious changes in brightness,missing target features and overlapping noise due to occlusion,and few rice data sets in the field environment with poor quality,an improved ResNet50 algorithm is proposed for rice disease identification in the field environment and the recognition system is designed.Based on the traditional ResNet50 algorithm,the transfer learning technique is adoptedto transfer the learning knowledge across domains to alleviate the overfitting phenomenon caused by insufficient and unbalanced data sets.The Cutout enhancement method is utilized to filter the feature information immediately to simulate the complex field environment and enhancethe generalization ability of the algorithm.Cosine annealing optimization strategy is adopted for the learning rate to improve the stability of the algorithm.The results show that the improved ResNet50 algorithmhas a recognition accuracy of 97.24%on a small rice disease dataset,which is significantly higher than that of the traditional ResNet50 algorithm,and the improved method also has an enhancementeffect on other convolutional neural network algorithms such as VGG16,GoogLeNet and MobileNetV3-large.The model is deployed in the system,which can provide technical reference for the development of rice disease identification in practical application engineering.

关 键 词：水稻 病害识别 Cutout增强 迁移学习 ResNet50 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]