An Integrated Deep Learning Approach for Crack Detection and Localization in Tyre Images Muhammed Sufail M K Centre for Artificial Intelligence TKM College of Engineering Kollam, Kerala, India sufailmuhammed074@gmail.com Chinnu Jacob Centre for Artificial Intelligence TKM College of Engineering Kollam, Kerala, India chinnujacob@tkmce.ac.in Christy D Ponnan Centre for Artificial Intelligence TKM College of Engineering Kollam, Kerala, India christyd@tkmce.ac.in Abstract—The production of tyres plays a crucial role in the expansion of the automotive sector. The journey from the first pneumatic tyre to the development of high-performance tyres has been a gradual process for the auto industry. Despite the extensive integration of artificial intelligence in manufacturing and production, the identification of cracks and defects in tyres still heavily relies on human inspection. This manual process is time-consuming and prone to inaccuracies, adversely affecting overall output. This study addresses these challenges by employing ResNet50 for the classification of normal and cracked tyres, and the YOLO (You Only Look Once) object detection algorithm for crack detection and localization. The utilization of ResNet50 and YOLO in this study offers a transformative solution to the challenges plaguing the tyre production sector. By harnessing ResNet50 for the classification of normal and cracked tyres, and integrating YOLO for precise crack detection and localization, this research significantly enhances the effi- ciency and accuracy of tyre inspection processes. With ResNet50 achieving an impressive average accuracy of 92.65% in tyre image classification and YOLOv5 demonstrating a remarkable mean average precision (mAP) of 94.95%, this composition not only streamlines operations but also elevates quality control standards. This technological advancement promises to reduce manual labor, mitigate inaccuracies, and ultimately bolster the automotive industry’s overall output and reliability. Keywords—Deep learning, tyre dataset, crack detection, YOLO, ResNet50 I. I NTRODUCTION The significance of tyre manufacturing in the automotive industry is highly noteworthy. The evolution of tyre pro- duction within the automotive sector has been a captivating journey marked by numerous technological advancements. From the inception of the first pneumatic tyre to the ongoing development of high-performance tyres, the tyre industry has made substantial progress and continues to innovate [1]. Due to the complex technology and materials involved in tyre manufacturing, tyres are susceptible to flaws such as cracks, bulges, cuts, and contamination from external objects [2]. Traditional visual inspection methods have limitations in accurately detecting minute flaws. To address these issues, we propose an automated approach using state-of-the-art deep learning models that can categorize and locate tyre faults. The primary cracks that can occur during tyre manufacturing are classified as ’chunking,’ ’chipping,’ and ’abrasion’ [3]. This study is focused on utilizing Resnet50 and YOLOv5 to classify and localize cracks within tyres. Within the tyre manufacturing sector, an effective model holds the potential to enhance both quality and time efficiency by significantly reducing the time required for fracture detection. Rapid identification of errors and defects within a constrained timeframe can positively im- pact the overall time efficiency of vehicle manufacturing.The key features of this study include: • A unified model for categorizing and identifying cracks in tyre images. • The model’s ability to identify even minor cracks within the tyre image. • The model performs excellently in terms of various performance measures such as precision, recall, and mean average precision (mAP). II. RELATED WORKS Several studies have contributed to the norms of providing classification and detection tasks in related areas. Cracks detection in tunnels and classification of defects in tyres are some related areas of work that contributed various deep learning algorithms to perform the operations. However, in the case of specific areas of tyre crack classification and detection, the number of authorized works is very limited. Kinasih et al. proposed a two-stage multiple object detection using CNN and Correlative Filter [4]. This method was used for object detection and witnessed substantial advancements, with the introduction of a two-stage multiple object detec- tion technique that leverages Convolutional Neural Networks (CNNs) and Correlative Filters. In this innovative approach, a Convolutional Neural Network is employed to extract dis- criminative features from input images and generate object proposals in the first stage. This initial step effectively reduces the search space, enhancing the efficiency of the subse- quent stages. The study conducted a thorough comparative analysis, pitting this novel two-stage method against well- established object detection algorithms, including Faster R- CNN, YOLOv3, and SSD. It produced better performance compared with the existing works the that particular area. A study by Xu et al. proposed an automatic defect detection and segmentation technique using a modified Mask R-CNN 979-8-3503-8168-9/24/$31.00 ©2024 IEEE 2024 IEEE Recent Advances in Intelligent Computational Systems (RAICS) | 979-8-3503-8168-9/24/$31.00 ©2024 IEEE | DOI: 10.1109/RAICS61201.2024.10690007 Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 09,2024 at 03:15:10 UTC from IEEE Xplore. Restrictions apply.