Improved YOLOv8 for substation image tampering detection

In modern power systems, the integrity and authenticity of image data are critical. However, image tampering can lead to misjudgments and erroneous decisions, threatening the safe and stable operation of substation. To address this issue, we propose an improved YOLOv8 model specifically designed for efficiently detecting tampered areas in power images. By introducing the Extra Deepwise Convolution(ExtraDW) module, optimizing the Spatial Pyramid Pooling module, and proposing the Cross Stage Featrue Pyramid(CSFP) module, we have enhanced the model’s accuracy in detecting tampered regions in complex backgrounds and improved the detection speed to meet the real-time requirements of power scenarios. Experimental results show that the proposed model performs excellently in all performance metrics, achieving an ideal balance of high precision and accuracy. The model’s speed is 96.49 FPS higher than the best existing model with similar detection accuracy, demonstrating its outstanding performance in substation scenarios and its capability to meet the demands of efficient real-time detection.