MMRD-DETR: An RGB–Depth Multimodal Fusion Framework for Defect Detection of Transmission Line Conductors and Ground Wires

Defect detection in UAV-based inspection of transmission line conductors and ground wires often suffers from limited accuracy and reduced robustness when relying solely on single-modal imagery. To overcome these challenges, this paper proposes MMRD-DETR, a multimodal defect detection framework centered on a Dual-Stream Wavelet Fusion Backbone. The proposed backbone performs frequency-domain decomposition and structured cross-modal alignment between RGB and depth features at the feature extraction stage, enabling effective integration of geometric structural information and texture cues. A dual-modal adaptive interactive encoder is designed to enhance the model's capability in processing dual-modal features and enhance stability. Furthermore, an Interact-Clean-boost fusion model is developed as the fusion layer to further improve model accuracy. Experimental results demonstrate that MMRD-DETR achieves a 4.8% improvement in mAP50 and a 4.5% improvement in overall mAP compared to the original single-modal model on conductor and ground wires defect datasets under challenging conditions. This algorithm provides an efficient and robust solution for multimodal defect detection in conductor and ground wires inspection.