Image Based Clothes Pattern Recognition Using Deep Neural Networks

With the rapid development of the clothing industry and the improvement of consumers’ demand for fashion taste,clothing image recognition technology has become a hot research topic and is widely used in e-commerce image search,intelligent clothing collocation,and customization.At the same time,the fashion version plays a crucial role in fashion design,directly affecting the clothing’s cut,line,and shape,determining comfort and market competitiveness.However,the traditional pattern design method has errors and limitations,which can not meet the modern fashion industry’s requirements for the pattern’s accuracy and efficiency.Therefore,pattern recognition technology has gradually become a research hotspot and is widely used in clothing production,sales and marketing,and other fields.It has a vast application prospect.Through pattern recognition technology,consumers can better understand clothing information and make more accurate choices.It can also assist clothing designers in creating clothing that better meets consumer demands,thereby increasing brand loyalty.Therefore,pattern recognition technology is a cutting-edge technology that can help the clothing industry better respond to market changes and consumer demands.This article will explain the research on deep learning-based clothing pattern recognition technology in three parts.(1)A high-quality dataset is needed to improve the accuracy and precision of clothing pattern recognition.This article contributes a dataset that includes images of clothing patterns and corresponding annotation information for eight categories,styles,sizes,and materials of clothing: T-shirts,Polos,shirts,fleeces,cardigans,suits,dresses,and windbreakers.This helps machine learning algorithms and models better understand and identify clothing patterns.Using a self-built dataset can improve the performance of machine learning algorithms and models,thereby improving the quality of clothing pattern recognition.(2)To explore the research on the classification and recognition of clothing patterns by existing algorithms,this thesis selected PCA,SVM,Le Net-5,Alex Net,and VGG16,respectively,from the traditional field of machine learning and deep learning for comparative analysis and research,to explore the effect,advantages,and disadvantages of different algorithms on the recognition of clothing pattern.Experimental results show that the VGG algorithm has the best performance and accuracy,recall rate is more than 81%,F1 score is 0.816,and accuracy is 95.40%.(3)To improve the recognition effect of clothing patterns,this thesis proposes an improved Mask RCNN network.To further enhance the recognition performance,this thesis makes improvements based on Mask RCNN.Firstly,combined with Cascade RCNN’s cascade head,a cascade detection head is added to Mask RCNN.To further improve the performance of the detector.Secondly,the depth separable convolution is combined with Mask RCNN,which can be used to replace the standard convolution layer in Mask RCNN.This will reduce the number of parameters and calculation costs of the model and improve the accuracy and speed of the model.Finally,the attention mechanism SENet is combined.It is used as a part of Mask RCNN to adjust the weight of each channel in the network to capture the feature information better and improve the model’s perception ability and detection accuracy.Experimental verification,the improved Mask RCNN network was3.6% higher than the primary Mask RCNN network in Bbox＿AP＿0.5 and 3.9% higher than the primary Mask RCNN network.The superior performance of the improved Mask RCNN algorithm in garment image recognition is verified,especially in segmentation,and accuracy is improved.