A comparative study of YOLOv5 and YOLOv8 for corrosion segmentation tasks in metal surfaces

Abstract

This study delves into the comparative efficacy of YOLOv5 and YOLOv8 in corrosion segmentation tasks. We employed three unique datasets, comprising 4942, 5501, and 6136 images, aiming to thoroughly evaluate the models’ adaptability and robustness in diverse scenarios. The assessment metrics included precision, recall, F1-score, and mean average precision. Furthermore, graphical tests offered a visual perspective on the segmentation capabilities of each architecture. Our results highlight YOLOv8’s superior speed and segmentation accuracy across datasets, further corroborated by graphical evaluations. These visual assessments were instrumental in emphasizing YOLOv8’s proficiency in handling complex corroded surfaces. However, in the largest dataset, both models encountered challenges, particularly with overlapping bounding boxes. YOLOv5 notably lagged, struggling to achieve the performance standards set by YOLOv8, especially with irregular corroded surfaces. In conclusion, our findings underscore YOLOv8’s enhanced capabilities, establishing it as a preferable choice for real-world corrosion detection tasks. This research thus offers invaluable insights, poised to redefine corrosion management strategies and guide future explorations in corrosion identification.

Keywords

• Corrosion
• Instance segmentation
• Image segmentation
• Computer vision
• Deep learning
• YOLO