A Real Time Performance Comparison of Rice Plant Disease Identification System using Deep CNN Models

Abstract

India is the second largest producer of the world and the largest exporter of rice. In order to ensure the healthy rice production, it is important to detect the diseases at early stage. Many approaches were proposed to solve the rice plant disease identification and it is identified through literature that those models were not given expected accuracy. In this study, an attempt has been made to determine the optimal suitable model among the four deep learning CNN algorithms to classify the rice leaf diseases. In this research study, 1600 images were used to classify into four class models: Healthy, brown spot, Hispa and leaf blast. From the results, performance comparison has been analyzed in terms of learning rate, precision and disease recognition accuracy. The deep learning CNN models, ResNet50, VGG19, InceptionV3, and ResNet152 have reached disease recognition accuracy of 75.76%, 87.64, 96.46% and 98.36%. The classification performance result demonstrates that Reset152V2 is the optimal CNN model can be used to classify of rice diseases.