Plant leaf disease classification using Wide Residual Networks

Most plant diseases have clear symptoms that can be recognized and diagnosed with the naked eye by an experienced plant pathologist. The process of diagnosing a disease with the naked eye is manual and slow, and its success rate depends on the ability of the pathologist. Today, many machine learning (ML) models are used to detect and classify plant diseases. In this article, an effort is made to apply the WRN (Wide Residual Networks) model in the field of plant disease classification. This model was trained to perform this classification task. WRN training is promoted using a transfer learning approach. In addition, a detailed experimental study of the WRN for the plant disease classification task on a PlantVillage test image set that includes 55,480 images is presented. The choice of WRN was mainly due to its enormous potential for image classification for various databases and its homogeneous structure. The WRN model represents a shallowest network with also a shorter training time and improved accuracy compared to the residual network (ResNet). The proposed WRN model achieved an accuracy of 99.9611% on a test set, illustrating the viability of the proposed model. The results obtained from the test showed that the model achieved the highest values compared to other deep learning models in the PlantVillage datasets. Overall, the process of training WRN models provides a robust way to classify plant diseases using automated networks on a huge global scale.