基于注意力机制与残差网络的玉米病害识别  

Maize disease recognition based on attention mechanism and residual network

作　　者：代丽[1] 倪光亮 Dai Li;Ni Guangiang(School of Economics and Management,Zhejiang Sci-Tech University,Hangzhou,Zhejiang 310018,China)

机构地区：[1]浙江理工大学经济管理学院,浙江杭州310018

出　　处：《计算机时代》2023年第8期84-88,共5页Computer Era

基　　金：国家自然科学基金(32071909);浙江省自然科学基金(LGN20E050006);杭州市科技发展计划项目(2020ZDSJ0488)。

摘　　要：为了解决传统玉米病害识别模型识别准确率低、收敛速度较慢、识别耗时长等问题,提出一种基于注意力机制与残差网络的玉米病害识别方法。使用批归一化(BN)加速模型的收敛,提高模型的泛化能力;将捷径连接应用于两个由残差模块与注意力机制SENet组合而成SE-ResNet结构上,来提高模型对特征的复用能力和提取能力。实验结果表明,所提模型的识别准确率可达99.08%,收敛速度更快,识别耗时更短,可以为玉米病害的实时检测提供技术支持。In order to solve the problems of low recognition accuracy,slow convergence and long recognition time of traditional maize disease recognition models,a maize disease recognition method based on attention mechanism and residual network is proposed.Using batch normalization(BN),the convergence of the model is accelerated to improve the generalization ability of the model.The shortcut connections are applied to two SE-ResNet structures made by combining the residual module and the attention mechanism SENet to improve the feature reuse and extraction ability of the model.The experimental results show that the recognition accuracy of the proposed model can reach 99.08%,and the convergence speed is faster,which can provide technical support for the real-time detection of maize diseases.

关 键 词：玉米病害 图像识别 卷积神经网络 注意力机制 残差网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]