A Universal Tire Detection Method Based on Improved YOLOv8

Abstract

Driving safety has become one of the top concerns of people in contemporary society, with regular tire inspection being indispensable to ensure safe driving. However, traditional methods of tire defect detection have encountered problems such as slow detection speed, complex tire defect backgrounds, and limited hardware resources. To address the above problems, this paper proposes a lightweight YOLOv8n-SOI algorithm for tire defect detection. First, a similarity-based attention mechanism (SimAM) was introduced to the C2f block of the backbone network to improve the ability to extract the shape features of irregular tire defects in complicated backdrops. Subsequently, four network structures were created in response to the need for lightweighting the detection procedure by incorporating the omni-dimensional dynamic convolution (ODConv) network structure at various positions. The network structure with the highest detection accuracy and smaller model parameters was chosen. Finally, Inner-intersection over union (Inner-IOU), an auxiliary edge loss function that concentrated more on the centre of the target, was added to the modified network to speed up tire defect regression convergence. Experimental results demonstrated that YOLOv8n-SOI surpassed the YOLOv8n algorithm regarding precision, recall, and mean average precision (mAP) by 3.8%, 1.1%, and 3%, respectively. Additionally, YOLOv8n-SOI reduced floating-point operations (FLOPs) by 6% and had a model size of only 6.5MB. This article presents the latest you only look once (YOLO) model for tire identification, contributing to the advancement of research in tire defect detection and providing a level of assurance for driving safety.

Keywords

• YOLOv8
• tire defect detection
• attention mechanism
• light weight
• deep learning