A novel deep learning-based hierarchical attention feature fusion network for automated detection of rice leaf diseases

Abstract

Rice, as a staple crop, plays a crucial role in global food security, especially in developing countries. However, rice production is significantly impacted by diseases such as Brown Spot, Hispa, and Leaf Blast, which can reduce crop yield. Traditional methods of disease detection rely on manual inspection, which is time-consuming, labour-intensive, and prone to errors. To address these challenges, this paper presents a novel deep learning-based model for automated rice leaf disease detection. This paper proposes a novel deep learning-based model, the hierarchical attention feature fusion network (HAFFN), designed to enhance rice leaf disease detection accuracy by addressing key limitations in existing methods. The HAFFN model integrates multi-level feature extraction with a hierarchical attention mechanism to improve the detection of both small and large infected areas. The core novelty of the proposed approach lies in the combination of the deep multiscale feature fusion network (DMFN), the adaptive multiscale feature aggregator (AMFA), and the deep hierarchical attention module (DHAM). The model was trained and tested on a publicly available rice leaf disease dataset and demonstrated superior performance compared to benchmark models like LeafNet, Xception, and MobileNetV2.