Crop Disease Detection Based on Unetmamba with Hierarchical Feature Fusion

Segmentation of crop disease images by deep learning technology can effectively assist ophthalmologists in crop disease screening, classification and monitoring of treatment progress. However, different crop disease leaf images have high inter-class similarity in scale, shape, position and texture, which brings great challenges to the automatic segmentation of crop disease images. A new image segmentation method for crop disease leaves based on U-Net and Mamba multi-feature fusion network (UNetMamba) is proposed. The network mainly includes several modules, i.e., Visual State Space (VSS), Selective edge clustering (SEC), progressive feature fusion module (GCF) and dynamic attention (DA). By selectively aggregating crop disease edge features and semantic information, SEC module refines the edge contour of the disease leaf image and recalibrates the position of the disease leaf image. The results on the crop disease image dataset validate that the proposed UNetMamba has great application potential in crop disease detection practice.