摘要
为了及时准确地识别和监测番茄病害,通过Flask框架开发一种基于改进Light-ResNet的番茄病害网页系统,系统使用预训练ResNet50模型作为基础网络,通过添加注意力机制、深度可分离卷积实现了ResNet50网络的轻量化改进及识别精度优化,并对其进行微调以适应番茄病害识别任务。最后通过将最终模型Light-ResNet50与传统ResNet50网络相对比,结果表明模型参数量缩减了39.84%,最终精度为97.27%,该系统具有更高的准确性和鲁棒性,为番茄生产提供了可靠的决策支持工具。
In order to timely and accurately identify and monitor tomato diseases,a tomato disease Web system based on improved Light-Res Net is developed using the Flask framework.The system uses a pre trained ResNet50 model as the basic network,and achieves lightweight improvement and recognition accuracy optimization of the ResNet50 network by adding Attention Mechanism and Depthwise Separable Convolutions.It is also fine tuned to adapt to the tomato disease recognition task.Finally,by comparing the final model Light-ResNet50 with the traditional ResNet50 network,the results show that the model parameter quantity is reduced by 39.84%,and the final accuracy is 97.27%.The system has higher accuracy and robustness,providing a reliable decision support tool for tomato production.
作者
林祺烨
王增宇
王润泽
LIN Qiye;WANG Zengyu;WANG Runze(Jilin Agricultural University,Changchun 130118,China)
出处
《现代信息科技》
2024年第14期49-53,58,共6页
Modern Information Technology
基金
吉林省大学生创新创业训练计划项目(S202310193051)。
