Enhancing Pakistan Rice Plant Disease Detection: A Highly Effective Pre-Trained CNN Model

Rice, a crucial global food source, faces numerous challenges impacting its quality and quantity. Pakistan is a significant rice producer and needs more agricultural technological advancements. Conventional disease detection methods for rice plants are time-consuming and ineffective. However, recent progress in agricultural technology, particularly machine learning & deep learning, holds great promise for improved disease identification. In this study, different datasets were considered for preparing a dataset that is specifically related to Pakistan’s environmental rice diseases. This study uses a prepared dataset to train the proposed fine-tuned ResNet101V2 model and for the robustness of the proposed model comparison with ResNet101, ResNet50V2, and DenseNet121 also performed in this study. Among the models evaluated, ResNet101V2 exhibited superior performance with a test accuracy of 81%, demonstrating high accuracy in identifying six distinct classes, including five rice leaf diseases (Leaf Blight, Leaf Smut, Hispa, Leaf Blast, and Brown Spot) and one healthy category specific to Pakistan environments. This study represents a significant advancement toward developing more efficient disease detection and management systems in rice farming, ultimately leading to reduced financial losses and improved agricultural practices.