Research on Power Inspection Image Recognition Model Based on Improved YOLOv10

In order to overcome the inefficiency and possible misjudgment problems faced by traditional manual review of transmission line inspection photos, this paper improves the deep learning vision model YOLOv10 from two aspects of data enhancement strategy and optimization loss function. Data enhancement strategy includes HSV, Mosaic, Mixup and similar enhancement. The loss function adopts positive and negative term loss function, and introduces three aspects: positive factor, negative factor and weight coefficient. The accuracy rate of the improved YOLOv10 model in the abnormal detection scenarios of power inspection such as balloons, kites, garbage, bird's nests, damaged insulators, flashover, and dropped lines has been greatly improved, thus comprehensively improving the overall efficiency of inspection operations, and providing important help for promoting the transformation of inspection work to automation and unmanned.