The tire factory mainly inspects tire quality through X-ray images. In this paper, an end-to-end lightweight semantic segmentation network is proposed to realize the error detection of bead toe. In the network, firstly, the texture feature of different regions of tire is extracted by an encoder. Then, we introduce a decoder to fuse the output feature of the encoder. As the dimension of the feature maps is reduced, the positions of bead toe in the X-ray image have been recorded. When evaluating the final segmentation effect, we propose a local mIoU(L-mIoU) index. The segmentation accuracy and reasoning speed of the network are verified on the tire X-ray image set. Specifically, for 512 × 512 input images, we achieve 97.1% mIoU and 92.4% L-mIoU. Alternatively, the bead toe coordinates are calculated using only 1.0 s.
Keywords: auxiliary supervision; defect detection; image processing; lightweight semantic segmentation network; tire X-ray images.