The Hull Structure and Defect Detection Based on Improved YOLOv5 for Mobile Platform

Hull inspection is of great significance to ensure the safety of ships for ocean transportation. Improving the intelligent performance of inspection can benefit the efficiency of hull inspection. In this paper, we propose a detection method for hull structure and defect using improved YOLOv5. Residual connections and weighted feature fusion are adopted to learn the importance of different input features, which enhance feature expression and improve the efficiency. Coupled with transformer block, we further capture global information and abundant contextual information. For the convenience of detection, we convert the detection model and deploy it on mobile devices. The detection model is packaged into apk, which is installed on the mobile platform to realize the hull image detection. Extensive experiments show that our improved YOLOv5 achieves better performance than the original algorithm on our hull structure and defect datasets. The results of mAP are 82.5% and 65.8%, which are improved by 3.7% and 2.8% respectively. Based on mobile deployment, the high accuracy of mobile detection demonstrates that our work is of great significance for the research of ship inspection.