Insulator Defect Detection Algorithm Based on RGB-D Multimodal Images

With the continuous increase in power demand, intelligent operation and maintenance of power grids has become crucial. As a key component of the power grid, the continuous inspection of transmission lines is vital for ensuring power supply safety. Insulators are commonly used critical electrical components in high-voltage transmission lines, which are used to achieve insulation between overhead lines, poles and the ground. Exposed to complex natural and electromagnetic environments for a long time, insulators are prone to damage, which can affect the safe and stable operation of transmission lines. Traditional manual inspection methods, due to their high cost and low efficiency, are gradually being replaced by drone inspections. However, accurately detecting defects in transmission components during drone aerial inspections remains a challenge. This study introduces a multi-modal insulator defect detection network based on an improved YOLOv8 to enhance the accuracy of defect detection. Firstly, based on the YOLOv8 network model, this paper utilizes an improved symmetrical backbone network structure to extract effective feature layers of RGB and depth features, and introduces an RGB-D data feature fusion module to achieve the fusion of RGB features and depth characteristics in spatial and channel positions. Moreover, to address the issues of heavy detection algorithm models and high computational power consumption, the YOLOv8 backbone network is modified and replaced with the lightweight MobileNetv3. Experiments on a private transmission dataset demonstrate that our model achieves an accuracy of 95.36% on the complex insulator dataset, significantly outperforming other advanced object detection methods, especially in scenarios involving small and overlapping targets, where the model maintains strong performance. This research proposes a simple yet effective multi-modal fusion model, which not only improves detection accuracy but also demonstrates excellent generalization ability, providing theoretical and practical significance for advancing defect detection in complex transmission scenarios.