Defect Inspection in Tire Radiographic Image Using Concise Semantic Segmentation

Abstract

Automated tire visual inspection plays an extraordinary important role in ensuring tire quality and driving safety. Due to the anisotropic complex multi texture and defect diversity characteristic of tire radiographic image, tire intelligent visual inspection has become one of the technical bottlenecks of intelligent manufacturing. In this work, a novel tire defect detection model using Concise Semantic Segmentation Network (Concise-SSN) is investigated for automated tire visual inspection. We perform an end-to-end pixel-wise tire defect detection by combining the power of an optimized semantic segmentation network and a compact convolutional neural network for classification. It can achieve the end-to-end pixel-wise full class defect detection and classification. The experimental results show superior performance on defect segmentation and classification tasks compared to state-of-the-art models with smaller model size and faster computation. Comparative experiments indicated that our Concise-SSN achieves the mPA score of 85.13%, the mIoU score of 77.34% on our test set. The accuracy of defect classification is 96.5% on average. Finally, we show faster computation (0.132 seconds per image) with competitive results on our dataset, which can meet the needs of online tire detection.