Comparing CNN Models for Rice Disease Detection: ResNet50, VGG16, and MobileNetV3-Small

Abstract

The Oryza sativa (rice) plant is an important staple food source, especially in the Asian region. Rice production is often disrupted by diseases such as Brown Spot, Leaf Scald, Rice Blast, Rice Tungro, and Sheath Blight, which can reduce yield and crop quality. This research aims to classify rice plant diseases using a deep learning approach with Convolutional Neural Networks (CNN) architecture, namely ResNet50, VGG16, and MobileNetV3-Small. The dataset used is Rice Leaf Disease Classification which consists of 1305 images with five disease labels. The data is divided into training, validation, and testing sets with proportions of 70%, 15%, and 15%. The results showed that the MobileNetV3-Small model provided the best accuracy on the test data of 79%, while VGG16 achieved the validation accuracy of 78.84%. Based on these results, MobileNetV3-Small is considered the most superior model for rice disease classification. This research shows the great potential of applying deep learning in automatic rice disease detection.

Keywords

• rice, plant disease, cnn, resnet50, vgg16, mobilenetv3-small, deep learning