针对现有的深度学习模型在水稻叶片病害的识别中准确率较低的问题，本文提出一种基于图像超分辨率预处理和二次迁移学习的水稻病害识别方法，通过采用超分辨率重建技术，可以获得更高质量的图像，从而提升识别率;通过使用二次迁移学习技术，引入由Inception块构成的AW模块构建网络模型AW-Net(add width modules to the net-work model)，该方法增加了模型的网络宽度，可以有效缩小类内距离，扩大类间距离，实现了对水稻叶片病害区域特征的有效提取，提升识别率。实验结果表明，本文方法识别的准确率显著提升。
Rice Disease Recognition Model Based on Image Super-Resolution Processing and Two-Step Transfer Learning
Aiming at the problem that the existing deep learning model has low accuracy in the recognition of rice leaf dis-eases,in this article we propose a rice disease recognition method based on image super-resolution preprocessing and two-step transfer learning.With the use of super-resolution reconstruction technology,higher quality images can be obtained,thus improving the recognition rate.By using the two-step transfer learning technology,AW modules composed of Inception blocks is introduced to construct the network model AW-Net(Add width modules to the network model).This method in-creases the network width of the model,which can effectively reduce the intra-class distance,enlarge the inter-class distance,realize the effective extraction of the features of rice leaf disease regions and improve the recognition rate.The experimental results showed that the recognition accuracy of our proposed AW-Net model was significantly improved.
image super-resolutiontwo-step transfer learningVGG16inception moduleimage classification
万方数据
维普
