Classification of Map Printing Defects Based on Convolutional Neural Network

Convolutional neural network is a key research direction of printing defect recognition and classification. In order to study the convolutional neural network structure applicable to the classification of map printing defects, firstly, the image was preprocessed and the acquired defects image was amplified by means of rotation and cutting. The convolution neural network was trained by defects image training set, the network model was studied and the optimal parameters were determined. Finally, the classification effect was tested by defects image test set. The experimental results showed that the convolutional neural network designed in this paper has better recognition accuracy when the learning rate is 0.005 and the batch size is 128.