An End-to-End Web-Based System for Rice Leaf Disease Classification Using Deep Learning

The smart agriculture or intelligent farming is getting popularity day by day since the machine learning (ML)-based technologies are using as proven effective tools. Disease of rice plant leaf is one of the most common obstacles in the production of rice to meet the huge amount of demand all over the world. This paper represents a website framework using deep learning to classify three common rice leaf diseases: Bacterial Leaf Blight, Brown Spot, Leaf Smut, and also can identify the Healthy one. Moreover, it provides an insight comparison analysis in results (accuracy, training time, model size, and parameters) of different state of art methods using same dataset and we have chosen the best one among them for website development. All the used models (InceptionV3, MobileNetV2, VGG19, ResNet50, VGG16, and AlexNet) have been customized for faster operation and lower storage. Preprocessing comprises organizing images in a uniform manner in order to maximize accuracy. The models were trained using Bahri's dataset of 9600 images of rice leaves, validated with 2400 images and tested with 4000 images. MobileNetv2 and VGG16, out of all the models, had greater accuracy results (98.05% and 99.3%, respectively). However, other evaluation matrices such as speedy training, small model size, and lower parameters have made MobileNetV2 perfect for this study. In the developed website, for each uploading rice leaf image, it can identify the disease class using MobileNetV2 model.