基于Resnet50的西藏高原地区玉米病害识别系统

Corn Disease Recognition System in Tibet Plateau Based on Resnet50 Model

本文基于计算机视觉,采用PyTorch框架,Resnet50迁移学习,余弦退火学利率衰减算法,对数据集进行训练获得高识别率、高效率的模型;用MYSQL数据库和Pyside2搭建前端界面实现玉米四种病害八种状态的病害识别系统。测试表明:该系统对收集到的西藏高原地区玉米种植的4种病害及病害程度的识别准确率达到81%以上,可实现用户上传病害图片或使用摄像头进行实时识别病害,并显示出该病害的相关简介和防止方法等一系列信息。该系统能够满足农业在识别防治玉米病害方面的需求。

The article was based on computer vision, applying PyTorch framework, Resnet50 migration learning, cosine annealing to obtain a high recognition rate, high efficiency model by training the data set. A front-end interface was built with MySQL database and Pyside2 to realize a corn disease recognition system for four diseases and eight states of maize. Tests showed that the system reached an accuracy rate of more than 80% for the four kinds of diseases and disease levels collected from maize plants on the Tibet Plateau. Users can upload pictures of diseases or use the camera to identify diseases in real time, and it will display the introduction and control methods related to the disease. The system is capable to meet the needs of agriculture in identification and control of maize diseases.