RDB-YOLOv8n: Insulator defect detection based on improved lightweight YOLOv8n model

Insulator defect detection is pivotal for the reliable functioning of power transmission and distribution networks. This paper introduces an optimized lightweight model for insulator defect detection, RDB-YOLOv8n, which addresses the limitations of existing models including high parameter counts, extensive computational demands, slow detection speeds, low accuracy, and challenges in deployment to embedded terminals. First, the RDB-YOLOv8n model employs a novel lightweight module, C2f_RBE, in its Backbone architecture. This module replaces conventional Bottlenecks with RepViTBlocks and SE modules with EMA attention mechanisms, significantly enhancing detection efficiency and performance. Secondly, the Neck of the model incorporates the C2f_DWFB module, which substitutes Bottlenecks with FasterBlocks and introduces depth-wise separable convolutions (DWConv) over standard convolutions to ensure accuracy and robustness in complex environments. Additionally, the integration of a BiFPN structure within the Neck network further reduces the parameters and computational load of the model. while simultaneously improving feature fusion capabilities and detection efficiency. Experimental results show that the enhanced RDB-YOLOv8n model achieves a 41.2% reduction in parameters and a decrease in GFLOPs from 8.1 to 7.1, with a model size reduction of 39.1% and an increase in mAP(0.5) by 1.7%, meeting the requirement of real-time and efficient accurate detection of insulator defects.