Tire Defect Detection Based on Improved DeepLabv3+

Tire defect detection is an essential step to control product quality, and accurate segmentation is the key to product grading. However, due to the fact that the internal defects of tires need to be judged manually through X-ray imaging, as well as the complex background and various types of defects, the grading of defects is marked differently, and the accuracy is affected by subjective factors. An improved algorithm based on DeepLabv3+ is proposed in this paper to solve the above problems. Firstly, the high-level semantic and low-level position features of image is integrate by encoder-decoder structure, which can face any input size. Secondly, atrous spatial pyramid pooling module and lightweight mobilnetv2 backbone can be used to accurately segment defects of different shapes and sizes, and reduce edge calculation pressure. The experimental results show that the improved deeplabv 3+ can detect different defects, while the accuracy performance of model is improved to 81.42% MIoU.