Explainable Deep Learning Approaches to Defective Tire Detection

Tires must be visually inspected to guarantee vehicle safety due to the sharp rise in the number of persons utilizing transport vehicles. This domain of study is important for maintaining traffic safety, lowering accident rates, and avoiding tire failure related to financial losses. Several technologies have previously been put forth to offer quick tire fault identification. Therefore, this study aims to improve damaged tire detection using multiple CNN-based models by identifying the best one by comparing their various performance measures. The performance of these models: YOLOv9, ResNet50, VGG-19, MobileNet, and Hybrid are compared to find out the best for this multi-data classification. The results of this research, indicate that the Hybrid model outperforms the vision transformer and other models with an accuracy of 89%, F1 score of 0.89, Precision of 0.89, Recall of 0.89, and AUC of 0.95. Explainable AI techniques like CAM, LIME, and Convolutional Neural Networks (CNNs) offer interpretable insights into tire defects, increasing the efficiency and accuracy of defective tire identification. A dataset of 1856 is used to train these algorithms including 828 photos of good tires and 1028 images of damaged tires.