Plant disease recognition using multi-scale feature progressive enhancement and attentional selective fusion

Plant disease recognition from leaf images can support timely and cost-effective crop monitoring. Convolutional neural networks have achieved strong performance in this task, but their effectiveness often decreases when only a limited amount of training data is available. Under such conditions, the network may fail to capture subtle texture and structural changes associated with different diseases. To address this limitation, this study proposes a multi-scale progressive feature enhancement and selective attention fusion model for plant disease recognition. The proposed method first extracts complementary texture information from leaf images to reduce its dependence on large-scale training data. A multi-level feature extraction module then processes each texture representation to capture both fine local patterns and broader structural characteristics. Next, a progressive feature enhancement module strengthens disease-related information at different levels. Finally, a bidirectional attentive selective fusion module enables interaction among the multi-stream features and emphasizes the most informative texture and structural cues while suppressing less relevant information. The proposed model was evaluated on five standard plant disease datasets and compared with fourteen recent methods. The experimental results demonstrate its effectiveness and consistency across different datasets. Moreover, the model contains only 1.77 million trainable parameters, making it suitable for deployment on resource-constrained agricultural monitoring systems. Source code: https://github.com/KarnatiMOHAN/MFPE-ASF