Train navigation rapid initial alignment system based on dual-antenna assistance

The invention provides a train navigation rapid initial alignment system based on dual-antenna assistance, and relates to the field of industrial defect detection. According to the method, a certain random position of an input normal sample image is cut to obtain a rectangular area with a random size, defect types and shapes of industrial products are collected as priori knowledge, a rectangle is subjected to mask transformation to form an image patch similar to the shape of a corresponding product defect, the patch is randomly rotated by a certain angle, and the image patch is obtained. The method comprises the following steps: performing random color jitter, pasting a patch at a random position of an original image to obtain an analog defect image, inputting the analog defect image and a normal image into a ResNet-18 neural network together, fusing a self-supervised attention module (SSPCAB) into the neural network, and using periodic focus loss (CFL) as a loss function of the neural network to obtain an an