A performance evaluation of convolutional neural network architecture for classification of rice leaf disease

Abstract

Plant disease is a challenge in the agricultural sector, especially for rice production. Identifying diseases in rice leaves is the first step to wipe out and treat diseases to reduce crop failure. With the rapid development of the convolutional neural network (CNN), rice leaf disease can be recognized well without the help of an expert. In this research, the performance evaluation of CNN architecture will be carried out to analyze the classification of rice leaf disease images by classifying 5932 image data which are divided into 4 disease classes. The comparison of training data, validation, and testing are 60:20:20. Adam optimization with a learning rate of 0.0009 and softmax activation was used in this study. From the experimental results, the InceptionV3 and InceptionResnetV2 architectures got the best accuracy, namely 100%, ResNet50 and DenseNet201 got 99.83%, MobileNet 99.33%, and EfficientNetB3 90.14% accuracy.