YOLO-SCFormer: A Feature Fusion Model for Small Target Foreign Object Detection in Transmission Lines

The presence of foreign objects on transmission lines significantly impacts the safe operation of the power grid. These foreign objects on transmission lines pose challenges such as small target size, low resolution, and varying dimensions. To address this, this paper introduces a small object detection algorithm tailored for foreign objects on transmission lines. Our contributions encompass a small object detection strategy, introducing a specialized layer to enhance the detection of small targets. Additionally, we propose the SCFormer attention mechanism, which integrates both channel and spatial features to augment modeling capabilities. Furthermore, we integrate deeply separable convolution into the backbone network of the YOLOv8 model, resulting in notable improvements in efficiency and speed. Experimental results demonstrate that the proposed approach accurately detects foreign objects on transmission lines, achieving a
$mAP^{val}0.5$
accuracy of 99.5%. Compared to traditional object detection algorithms, our method strikes a balance between speed and accuracy, establishing itself as a state-of-the-art algorithm.