摘要
针对现有的多种植物病害分类识别方法存在准确率不高的问题,提出一种基于混合注意力机制深度残差网络的植物病害识别方法。该识别方法通过在传统的残差神经网络上,增加空间注意力模块以及通道注意力模块,使网络对于图片的细节部分的注意力更高,提取的特征更为丰富。为了验证改进后的网络模型对植物病害分类识别的有效性,使用改进的模型与现有方法进行对比试验。结果表明,基于混合注意力机制的残差网络对于植物病害有很好的识别效果,实验识别准确率达到92.08%,该方法可为植物病害的分类识别提供参考。
For the existing plant diseases classification and recognition algorithm’s accuracy is not high,we post a new algorithm of mixed attention deep residual network plant diseases clarification and recognition.The recognition algorithm aggrandizes spatial attention module and channel attention module to the traditional residual neural network,which makes the network have higher attention to the details of pictures for extracting more features.In order to verify effectiveness of plant-disease classifications and recognitions from improved network models,it made a series of comparative experiments between improved models and the existing methods.The experimental shows that,mixed attention algorithm’s more accurate than the existing algorithm and have a certain degree of improvement,the experimental recognition accuracy is up to 92.08%.This method can provide certain references for subsequent classifications and recognitions of plant diseases.
作者
尚远航
余游江
吴刚
SHANG Yuanhang;YU Youjiang;WU Gang(College of Information Engineering,Tarim University,Alar,Xinjiang 843300;Modern Agricultural Engineering Key Laboratory at Universities of Education Department of Xinjiang Uygur Autonomous Region.Alar,Xinjiang 843300)
出处
《塔里木大学学报》
2021年第2期94-103,共10页
Journal of Tarim University
基金
新疆生产建设兵团科技攻关与成果转化项目“基于小样本数据的新疆红枣缩果病预测技术研究”(2015AC023)。
关键词
植物
病害
图像处理
通道注意力
空间注意力
plants
diseases
image processing
channel attention
spatial attention