Exploiting The Strength of Modified Parrot Optimization Algorithm for Enhancing Rice Leaf Disease Detection Using Convolutional Neural Network and Transfer Learning

Timely and accurate identification of rice leaf diseases is critical for optimizing crop productivity and safeguarding global food security. This study developed an innovative deep learning framework that incorporates the DenseNet121 architecture, optimized through a modified Parrot Optimization Algorithm (POA), to achieve precise classification of rice leaf diseases. The modified POA, an enhanced variant of the original algorithm, integrates Mutation random opposition-based learning (mROB) and Brownian motion mechanisms to improve optimization efficiency. The proposed model demonstrates superior performance by effectively tuning critical hyperparameters, including batch size, learning rate, dropout rate, and the number of neurons. Evaluations conducted on the RLD dataset revealed that the modified POA-DenseNet121 model outperformed established pretrained models, such as VGG19, DenseNet201, InceptionV3, EfficientNetB0, and ResNet50. The proposed model achieved remarkable performance metrics, including 98.5% accuracy, 98.6% precision, 98.4% recall, and 98.5% F-measure. Furthermore, the application of optimization strategies, including step decay learning schedules and early stopping, enhanced the model’s robustness and minimized the risk of overfitting. This study underscores the potential of the modified POA-DenseNet121 framework as a scalable and efficient tool for advancing agricultural diagnostics and addressing challenges in rice disease management.