Improved YOLO-Based Corrosion Detection and Coating Performance Evaluation Under Marine Exposure in Zhoushan, China

In response to the challenges of metal corrosion detection and anti-corrosion coating performance evaluation in the marine environment of Zhoushan, this study proposes an improved object detection model, YOLO v5-EfficientViT-NWD-CCA, to enhance the recognition accuracy and detection efficiency of corrosion images on marine structures. Based on YOLO v5, the model incorporates the EfficientViT backbone network, NWD (Normalized Wasserstein Distance) loss function, and CCA (Criss-Cross Attention) attention mechanism, outperforming comparative models across multiple key metrics. Experimental results show that the proposed model increases precision from 0.73 to 0.76 (approximately 4% improvement) and raises the True Positive rate from 0.66 to 0.70 (approximately 6% improvement) according to the confusion matrix, demonstrating more stable overall detection performance. Building on this, the study combines the model’s detection results to conduct a quantitative analysis of the corrosion area of eight types of anti-corrosion coatings in two typical marine environments—tidal zones and fully immersed zones—across different exposure periods (24, 60, and 96 months). The results indicate that the tidal zone presents a harsher corrosion environment, with corrosion severity significantly increasing over time. Fusion-bonded epoxy coatings, powder epoxy coatings, and fluorocarbon coatings exhibit good corrosion resistance, whereas chlorinated rubber coatings and conventional epoxy coatings perform poorly. This study not only achieves intelligent identification and precise quantification of corrosion areas but also provides a scientific basis for the selection and evaluation of anti-corrosion coatings in different marine environments.