Application of Convolutional Neural Network ResNet-50 V2 on Image Classification of Rice Plant Disease

Abstract

Rice is the most important crop in global food security and socioeconomic stability. A part of the world's population makes rice a food requirement but the problem is found that all rice varieties suffer from several diseases and pests. Therefore, it is necessary to ensure the quality of healthy and proper rice growth by detecting diseases present in rice plants and treatment of affected plants. In this study, the Convolutional Neural Network (CNN) algorithm was applied in classifying diseases on the leaves of rice plants by experimenting with several parameters and architecture to get the best accuracy. This study was conducted image classification of rice plant disease using CNN architecture ResNet-50V2 with data using preprocessing Augmentation. The test was conducted with three optimizers such as SGD, Adam, and RMSprop by combining various parameters, namely epoch, batch size, learning rate, and SGD and RMSprop optimizers. Division of image data with 70:30 ratio of training data and test data; 80:20; 90:10. From these results, it was found that Adam was the best optimizer in the 80:20 data division in this study with an accuracy level of 0.9992, followed by the SGD optimizer with an accuracy level of 0.9983, while the RMSProp optimizer was ranked third with an accuracy level of 0.9978.

Plant phenotyping relevance

イネ葉の病害状態を画像から分類するCNN手法が研究の中心であり、植物の病害表現型を直接推定しているため。

Code and data availability

The paper's rice leaf disease image classification uses a publicly available Kaggle dataset of rice leaf images (5932 images across Bacterial blight, Blast, Brown spot, Tungro), which is the paper-specific phenotyping image input. No author analysis code or trained model checkpoints are reported.