A Comprehensive Analysis of Multi-Modal Deep Transfer Learning for Rice Leaf Disease Detection

In the context of Bangladesh's agricultural landscape, rice cultivation constitutes a significant role in the nation's economic prosperity. However, several disease infestations in crops such as Bacterial Blight, Blast, and Brown Spot, impose substantial financial burdens on farmers annually. With rapid and accurate disease detection, farmers can minimize crop losses, optimize resource usage, and enhance the resilience and sustainability of their agricultural operations. To address these challenges and accelerate disease detection, this study investigates the potential of machine learning algorithms, specifically transfer learning models like VGG19, ResNet50, Inception V3, and Vision Transformer (ViT), in enhancing the accuracy and scalability of rice disease detection and classification. With a dataset of 4704 images captured from rice fields, the study ensures consistent model parameters and diverse evaluation metrics to identify the optimal framework for rice disease detection. Finally, the study concludes with a comprehensive analysis, identifying the most efficient and accurate transfer learning model for practical agricultural use among the proposed methods.