A Corrosion Detection Method for Hydraulic Metal Structures Based on an Improved YOLOv10 Model

To achieve corrosion detection of hydraulic metal structures based on deep learning, reduce subjectivity in the corrosion annotation process, and enhance annotation efficiency, this study proposes a corrosion detection method for hydraulic metal structures based on a YOLOv10 serial architecture that incorporates a Transformer module. The serial architecture first uses an object detection model to locate corrosion, followed by a classification model to assess corrosion severity. In terms of F1-score, the proposed corrosion detection network shows a 1% improvement over the baseline. For accuracy, after transfer learning, the proposed corrosion classification model outperforms the baseline by 8.8%. This method was implemented in real-world scenarios, successfully enabling corrosion detection and severity classification of hydraulic metal structures, and verifying the feasibility of the serial architecture.