基于迁移学习和残差网络的农作物病害识别

Crop Disease Recognition Based on Residual Network of Transfer Learning

摘要: 针对人工识别农作物病害受主观因素的影响较大，传统的农作物病害识别训练时间较长、训练中有过多参数的问题，本文提出一种基于迁移学习和残差网络模型对农作物病害进行识别的研究方法。本研究以常见的6种农作物共19类病害叶片图像为研究对象，对农作物病害图像进行识别。首先对图片数据集进行旋转翻折、随机裁剪等预处理扩充数据集，以减少网络模型的过拟合。基于扩充后的农作物病害数据集，本文使用Resnet-50网络模型，并利用迁移学习的方法对农作物病害进行识别，进行参数微调，最后对模型进行训练测试。试验结果表明，与传统的识别模型相比，该方法能够快速准确的对农作物病害进行识别，并且识别的准确率高达91.24%。

Abstract: Aiming at the problem that the artificial recognition of crop diseases is greatly affected by subjective factors, the traditional crop disease recognition training takes a long time and there are too many parameters in the training, this paper proposes a method based on migration learning and residual network model for crop diseases. In this study, we used images of 19 types of diseased leaves of 6 common crops as the research object to identify crop disease images. Firstly, the image data set is preprocessed to expand the data set such as rotation, folding, random cropping, etc., to reduce the overfitting of the network model. Based on the expanded crop disease data set, this paper uses the Resnet-50 network model, and uses the transfer learning method to identify crop diseases, fine-tune the parameters, and finally train and test the model. The test results show that compared with the traditional recognition model, this method can quickly and accurately recognize crop diseases, and the recognition accuracy rate is as high as 91.24%.