A Study of Corrosion-Grade Recognition on Metal Surfaces Based on Improved YOLOv8 Model

Typical metal equipment in substations is exposed to high-temperature, high-humidity, and high-salt environments for a long time, and surface corrosion is a serious threat to operational safety. Traditional manual inspection is limited by the complexity of the environment and subjective assessment errors, and there is an urgent need for a method that can quickly and accurately locate the corrosion area and assess the degree of corrosion. In this paper, based on YOLOv8, the feature extraction ability is improved by introducing the attention mechanism; a mixed-mixed-sample data augmentation algorithm is designed to increase the diversity of data; and a cosine annealing learning rate adjustment is adopted to improve the training efficiency. The corrosion process of metal materials is accelerated by a neutral salt spray test in order to collect corrosion samples at different stages and establish a dataset, and a model of a corrosion-state recognition algorithm for typical equipment in substations based on an improved YOLOv8 model is established. Finally, based on ablation experiments and comparison experiments, performance analyses of multiple algorithmic models are conducted for horizontal and vertical comparisons in order to verify the effectiveness of the improved method and the superiority of the models in this paper. The experiments verify that the improved model is comprehensively leading in multi-dimensional indicators: the mAP reaches 96.3% and the F1 score reaches 93.6%, which is significantly better than mainstream models such as Faster R-CNN, and provides a reliable technical solution for the intelligent inspection of substation equipment.