Automated Rice Disease Diagnosis using Deep Learning: A Wavelet-Enhanced VGG-16 Model with Manta Ray Optimization

Abstract

Plant diseases significantly impact crop yields and financial stability, especially in Asia where rice is a staple crop. This study introduces the Automated Deep Learning with Wavelet Neural Network (ADLWNN) model to effectively identify and classify rice plant diseases. The ADLWNN model integrates the VGG-16 Convolutional Neural Network (CNN) for feature extraction from rice plant images. VGG-16, with its deep architecture of thirteen convolutional layers and two fully connected layers, is fine-tuned for binary classification by reinitializing the final SoftMax layers. Hyperparameter tuning is achieved through the Manta Ray Foraging Optimization (MRFO) algorithm, which mimics the foraging behavior of manta rays using techniques like somersault foraging and cyclone foraging to optimize the model parameters.For robust recognition, the Wavelet Neural Network (WNN) is employed, which decomposes input signals into simpler wavelet components for precise pattern identification. The WNN's wavelet analysis, combined with the optimized features extracted by VGG-16, enhances the model's classification capability. Simulation results on a rice plant image dataset show that the ADLWNN model achieves a remarkable 98.18% accuracy, outperforming existing methods in sensitivity, specificity, precision, and F-score. This comprehensive approach demonstrates the ADLWNN model's effectiveness in automated rice disease diagnosis, offering a valuable tool for safeguarding crop yields.