Multi-Contrast Convolution Neural Network and Fast Feature Embedding for Multi-Class Tyre Defect Detection

With an increase in the computational power, there has been a rapid shift towards automation by using computer vision. One of the industries that have benefited the most from these technological advancements is the tyre industry. Under certain situations, even minor tyre defects can be dangerous and they must be detected at the earliest, with no or a small error of being missed. Statistics show that annually about 11,000 tire-related crashes will occur. Despite the hype, due to the increased accuracy in detecting these defects, the number of such accidents will also reduce. This paper proposes the two algorithms used for tyre defect detection. The algorithms used are the multi-contrast convolutional neural networks (MC-CNN) and convolution architecture for fast feature embedding. For experimental purposes, a novel dataset is proposed by collecting images from Google and the proposed algorithms are evaluated based on this dataset. The obtained results show that the MC-CNN algorithm projects superior performance in defects detection. Although the proposed model has been executed on car tyres, the same can br extrapolated into other domains such as aircraft tyre defect detection given with the appropriate dataset.