Detection of Violations in Power Production Operations Based on Improved Yolov8

To address the issues of insufficient feature extraction capabilities, high computational complexity, and slow detection speed in power production violation detection models, an improved YOLOv8 algorithm is proposed. Firstly, an SDWblock (Sobel depthwise block) is designed, integrating SobelConv and Conv branches to increase feature extraction capability for object boundary information, while further reducing the model's parameter count and computational complexity through DWConv. Secondly, an LSDH head (Lightweight shared detection head) is devised, which shares detection head parameters across multiple feature scales and utilizes different batch normalization layers to handle features at various scales, resulting in a significant reduction in the model's parameter count, maintaining high detection accuracy, and improving inference efficiency. Lastly, experimental results validate the effectiveness and superiority of the improved YOLOv8 algorithm in complex scenarios. Ablation experiments further confirm the effectiveness of SDWblock and LSDH head in the object detection model. The improved YOLOv8 algorithm enhances the accuracy and speed of power production violation detection, providing effective technical support for safety supervision in the power industry.