Complementary-aware Synergistic Fusion : Advancing Precise Wheat Lodging Mapping Using UAV-Borne RGB Imagery and DSM

MWLNet (Mapping Wheat Lodging Network) is a deep learning model designed for accurate mapping lodged wheat using multi-modal data, including RGB images and Digital Surface Models (DSM). The model utilizes a hybrid feature fusion mechanism to effectively handle complex scenes and accurately identify the boundaries of lodged wheat.

Key Features

• Hybrid Cross-Modal Fusion: MWLNet integrates RGB and DSM data through advanced feature fusion techniques, such as the Discrepancy Modality Recalibration (DMR) module and the Dynamic Feature Consolidation (DFC) module. These mechanisms enhance the complementary features of the two modalities while suppressing redundant information.

• Bi-directional Semantic Synchronization (BSS): At the core of the model is the Bi-directional Semantic Synchronization (BSS), which improves the model's ability to capture global and local dependencies between modalities. The Self-Cross Cooperative Attention (SCCA) mechanism within the BSS helps to effectively fuse intra-modal and cross-modal information.

• Attention Mechanisms: The model uses both channel-wise and spatial attention to adaptively emphasize important features while suppressing irrelevant or noisy information, ensuring improved performance in complex background scenarios.

Code Release

The complete code, along with pretrained models, will be publicly released after the paper is published. Stay tuned for further updates!