Abstract
Overhead transmission line hazard detection is related to the proper functioning of power communication systems and society. With the development of Unmanned Aerial Vehicles (UAVs) and deep learning, deep-learning-based hazard detection using UAVs has received extensive attention. Currently, research in this direction faces three main challenges: complex background interference, small-scale problems, and efficiency-performance balance. To address the above challenges, this study introduces Mamba based on State Space Models (SSMs) with linear complexity and proposes the UAV-MDMamba model for overhead transmission line hazard detection. We design a Multi-Directional Mamba (MDMamba) block to improve image spatial modeling and complex background suppression, which helps to capture hazardous areas in small-scale situations. Moreover, Patch-Level Inference Enhancement (PLIE) is designed to improve the detection accuracy of small targets in inference. Finally, we collect and label a dataset of overhead transmission line hazard detection for complex scenarios. Extensive experiments demonstrate that UAV-MDMamba performs excellently on the dataset. Therefore, this study improves the efficiency and accuracy of detecting hazards on overhead transmission lines.