Improve the spot-like coding detection of U-net auto-encoder

The dot-like spray code on the product packaging has always been a difficult problem in industrial inspection due to its complicated background in the printing area, diverse characters in the code, and changeable fonts. With the powerful capabilities of deep learning in the field of computer vision, related algorithms have become popular solutions in the field of computer vision in recent years. The method based on convolutional autoencoders to achieve inkjet detection on food packaging boxes has become a feasible solution. This method takes the mask of the coding character area on the food packaging box as the final goal of network learning. The network inputs the picture with the coding area, and reconstructs the original background of the coding area. At the same time the network uses the residual channel attention mechanism to restore the details of the image, and at the same time introduces the gaussian operator to calculate the loss of network reconstruction. Since the convolutional auto-encoder is an unsupervised learning method, the training data does not require a large amount of manual labeling, and a good solution is proposed for scenarios such as small data sets in industrial production and difficulty in labeling. Through real-time testing of the coding data of the pipeline, it is verified that the method can effectively detect the coding area.