Abstract:

To inherit and innovate the traditional blue calico vein patterns making use of the digital technology, the image elements constituting blue calico vein patterns were extracted to form a structured database for vein pattern elements. A classification method based on convolutional neural networks was proposed. The pattern elements were firstly extracted from the captured 128 images of blue calico to form an image sample database with a total of 21 212 images. Secondly, 80% of image samples in the database were randomly selected as the training set and the rest 20% as the testing set. The training samples were convoluted by a 5×5 convolutional kernel size, and the obtained feature maps were pooled. After computing through 3 convoluting layers, 3 pooling layers and 2 full connection layers, 12 classification types were obtained by using Softmax classifier. Eventually, the optimal network model parameters were acquired and ideal classification results were obtained through deep learning of the image element samples. The experimental results show that the 12 types of vein pattern elements of blue calico, produced by the convolutional neural networks model, are proved to be with an average accuracy of 99.61%, and detection accuracy of 98.5%. This work provides new ideas for studying blue calico vein patterns.

Key words: blue calico, classification of vein pattern elements, convolutional neural network, digital textile