Insulator Detection Based on Feature-Enhanced Convolution Attention Method

With the purpose of solving the problems of low inspection efficiency and high field inspection danger encountered in current insulator detection, in this paper we propose the Feature-Enhanced Convolution Attention YOLO (F-ECA YOLOv8), an improved YOLOv8 insulator detection algorithm for infrared images. Firstly, the ordinary convolution in C2 is substituted by the Depthwise Separable convolution, improving the target detection accuracy which enhance its adaptive ability and reducing the computational complexity. Next, the Fusion-SE-Coord Attention(F-SCA) module is designed to improve the model's ability to adapt to key features and to capture channel features and spatial features, which is able to fuse global and local information effectively. Finally, WIOU is used to replace the original loss function to further address the issue of the unbalance quality of the sample. Compared to the YOLOv8s,the F-ECA algorithm scored 0.5% and 6.7% higher in mAP@0.5 and mAP@0.5-0.95 separably, indicating that the algorithm can be effectively applied to detect insulators in infrared images in practical scenes.