Optimized Deep Learning and Hybrid Artificial Intelligence Framework for Rice Disease Detection based on Key Parameter Analysis

Abstract

Rice leaf diseases have a great impact on the productivity and food security of crops. Thus, there is a necessity for development of accurate and fast automatic rice leaf disease detection. In the current paper, we propose an optimized deep learning model for multi-class rice leaf disease detection through transfer learning and parameter optimization. We use a dataset that includes 3,829 images belonging to six classes (diseases and healthy-leaf). Images are resized, normalized and augmented through rotations, flips, and zooms. Pre-trained architectures ResNet-50 and EfficientNet-B3 are fine-tuned, and comparative experiments are carried out using accuracy, precision, recall, and F1-score metrics. Results show that the EfficientNet-B3 network outperforms the ResNet-50 model with 97.84%, 97.62%, 97.71%, and 97.66% of accuracy, precision, recall, and F1-score respectively. Our model demonstrates good performance for similar diseases like Leaf Blast and Leaf Scald.