YOLOv8-CAFMFusion for Image Segmentation of Substation Operation Elements in Complex Environments

This paper presents an enhanced segmentation method for substation operation elements, leveraging YOLOv8 integrated with the Content-Aware Fusion Module (CAFM) from HCANet. The proposed model aims to address the challenges of detecting safety violations in complex and dynamic substation environments, where objects often overlap or are occluded. By incorporating a content-guided attention fusion mechanism, the model captures both global and local contextual information, significantly improving pixel-level segmentation accuracy. Experimental results demonstrate that the YOLOv8-CAFMFusion model achieves a 3.2% improvement in mAP (from 79.3% to 82.5%) and a 17.2% reduction in inference time (from 29 ms to 24 ms). This enhanced performance enables more accurate and real-time detection of safety violations, such as improper use of protective gear or unauthorized access to restricted zones. The use of image segmentation for substation operation monitoring offers significant advantages over traditional object detection, including precise identification of object boundaries and better handling of occlusions, ultimately contributing to a safer and more efficient operational environment.