Detection and segmentation of overhead transmission line icing images via an improved YOLOv8-seg

Overhead transmission lines ice-covered endanger the stable operation of power systems seriously. Different types of icing have different hazards to the power grid. To ensure accurate detection and segmentation of different types of line icing, an improved YOLOv8-seg algorithm is used for icing detection and segmentation of overhead transmission lines. The algorithm introduces deformable convolution v2 (DCNv2) to the Cf2 module at the bottom of the backbone network to enhance the network’s ability to learn texture and edge features in images. To further improve the detection and segmentation accuracy of the model, the algorithm is improved by introducing the triplet attention mechanism into the network, improving the upsampling method of the network, and applying the slide weight function to the loss calculation. The experimental results demonstrate that the improved YOLOv8-seg algorithm achieves accuracies of 95.28% and 95.04%, with accuracies of Box_mAP@0.5 and Mask_mAP@0.5, respectively and the maximum and minimum difference in detection and segmentation accuracy for different types of icing should not exceed 3%. These results verified the effectiveness of the improved algorithm and underscored its performance advantages. The detection results can provide a reference for power grid icing disaster prevention and control and icing degree assessment.