Classification And Recognition Algorithm Of Embroidery Images Based On Deep Learning

Embroidery,as a traditional Chinese handicraft,is an indispensable part of intangible cultural heritage.In this paper,for the five types of embroidery images of Su embroidery,Shu embroidery,Yue embroidery,Xiang embroidery,and the unique Tu embroidery of Qinghai region,the deep learning method is used to extract the corresponding image features according to the different styles and patterns of each embroidery,and then carry out Classification and identification provide a basis for embroidery image research and digital protection.Convolutional Neural Networks(CNN),with its powerful image processing capabilities,are widely used in various image classification systems and have achieved remarkable results.In image recognition,Faster R-CNN detection algorithm based on RCNN has received more and more attention in recent years.Compared with traditional image processing algorithms,this type of algorithm can extract deeper pattern features in complex embroidery images,improving the robustness and recognition accuracy of the algorithm.The research results of this paper mainly reflect the following aspects:1.Traditional machine learning algorithms are not ideal for extracting the features of embroidery images,which makes the efficiency of embroidery image classification inefficient.This paper proposes to use the deep learning-based AlexNet network for image classification.In the AlexNet network,the ReLu activation function and local response normalization(LRN)are used.The Dropout layer is added to make the classification effect better.On this basis,pre-trained model parameters are loaded,which greatly improves the convergence speed of the model.2.In terms of image recognition,because the patterns in embroidery are too complicated and various,in order to extract deeper pattern features from the embroidery image,this paper uses Faster R-CNN network to recognize different types of embroidery patterns,and uses ResNet50 and VGG16 respectively Compare as a Faster R-CNN feature extraction network.The experimental results show that the pre-trained AlexNet model makes the accuracy of embroidery image classification significantly improved;meanwhile,the image recognition effect of Faster R-CNN based on ResNet50 is better.