The choice of deep learning methods for solving the problem of recognizing plant diseases for cases of a small training sample

Loss of yield due to plant diseases is a serious problem for rural farmers, the economy and food security, requiring timely measures to identify and prevent diseases. Recently, neural network methods of deep penetration have been successfully applied to solve the problem of recognizing plant diseases from photographs of their leaves. This study analyzes the methods used to train deep convolutional neural networks for cases of a small training set. For the PDD data (http://pdd.jinr.ru/crops.php), the transfer learning technique and the Siamese neural network method with a three-term error function were applied, which allowed achieving 99.5% of the classification accuracy. The reported study was funded by RFBR according to the research project № 18-07-0082.