Rice Disease Recognition Using Effective Deep Neural Networks

摘要

Rice is the most important grain in Thailand for both consuming and exporting. One of the critical problems in rice cultivation is rice diseases, which affects directly to the yield. Early disease recognition is handled by a human, which is difficult to achieve high accuracy and the performance depends on the farmer's experience. To overcome this problem, we did three folds of contributions. First, an infield rice diseases image dataset, named K5RD, was created. Second, a number of additional techniques to enhance the classification scores including data augmentations and learning rate adjustment strategies were carefully surveyed. Third, a set of selective deep learning models including ResNets and DenseNets were applied to classify such rice diseases. The experimental results reveal that the proposed framework can achieve high performance, which its F1 score is higher than 98% on average, and has the potential to be implemented as a practical system to provide to Thai farmers in the future.