基于迁移学习和轻量级卷积神经网络农作物病害识别

Agricultural Crop Disease Identification Based on Transfer Learning and Lightweight Convolution Neural Network

我国的农作物病害识别主要依靠人工,借助计算机进行农作物病害识别存在病害样本数据量少,识别的精度低、实时性差等问题。针对上述问题,建立了水稻、小麦、玉米三种作物的15种常见病害作为实验样本集,利用迁移学习的方法,在ImageNet数据集取得Shufflenet v2 1×、MobileNet v3_arge、GhostNet的预训练模型,迁移网络的所有权重信息进行训练。实验结果表明:基于迁移学习的MobileNet v3_age模型在农作物病害识别上所达到的效果最好,准确率为99.27%,模型大小仅4.02 M。

At present, the identification of agricultural diseases in China mainly depends on manual work. However, there are still some problems in the current identification of agricultural diseases with the help of computer, such as small amount of disease data, low accuracy of identification, poor real-time performance. In view of the above problems, 15 common diseases of rice, wheat and corn were taken as the experimental sample set, using the method of transfer learning. Obtain the pre-trained models of Shufflenet v2 1×, MobileNet v3_large, and GhostNet from the ImageNet data set. Migrate all weight information of the network for training, The experiment results show that MobileNet v3_large model base on transfer learning is the best in the recognition of agricultural crop diseases. The accuracy rate is 99.27% and the model size is only 4.02 M.