Research On Collar Image Classification Based On Convolutional Neural Network And Attention Mechanism

As people’s living standards improve,the demand for clothing quality is also increasing.In the fields of clothing production and sales,the evaluation of clothing quality is becoming increasingly important.Among them,the clothing collar,as an important part of clothing,directly affects the overall effect of clothing through its design and production quality.Therefore,the classification and evaluation of clothing collars is an important aspect of clothing quality evaluation.Traditional clothing collar classification methods are mainly based on manual feature extraction and classifier construction.However,the classification effect of this method is limited by the selection of features and the design of the classifier,resulting in low accuracy.In recent years,with the continuous development of deep learning and computer vision technology,research on image classification based on convolutional neural networks has made breakthrough progress,and many excellent image classification networks have emerged and been widely used in various fields.However,due to the complexity and diversity of clothing collar data,a single convolutional neural network model often cannot achieve good classification performance in collar classification tasks.Therefore,this paper summarizes the characteristics of collar image data and studies how to use deep learning technology to improve the accuracy and robustness of clothing collar classification.The specific research contents are as follows:1.This paper reviews the background and significance of collar image classification,including its application value in the clothing industry and the development status of related technologies.It summarizes the achievements of collar image classification methods based on deep learning,discusses the challenges of collar image classification tasks,and also looks forward to the possible development directions of collar image classification research based on deep learning in the future.2.In view of the problem that clothing collar image data has multiple objects and the recognition target area is small,a novel module called MFA(Multi-scale Feature Attention)is proposed.The module combines attention and multi-scale feature extraction methods to obtain more comprehensive and rich feature information on the one hand,and makes the network more focused on important features on the other hand.Based on the MFA module,an efficient clothing collar image classification deep convolutional neural network MFANet is proposed.Through multiple comparative experiments on collar classification dataset Collar6,general dataset CIFAR-10,and Deep Fashion6,with multiple classical and advanced convolutional neural networks,the network model proposed in this paper not only achieves superior performance in collar image classification tasks,but also has excellent generalization performance.Through ablation experiments,it is proved that the combination of attention mechanism and multi-scale features can effectively improve the recognition performance of network models on complex images.3.In view of the problem that traditional convolutional neural networks cannot effectively process local details and global contextual information in collar images in complex scenes,this paper proposes an S-MHSA(Small Multi-Head Self-Attention)module.The module uses local encoding units and downsampling to effectively reduce computational resources while retaining the ability of multi-head self-attention to model global contextual information.Based on S-MHSA,a new collar classification network MFA-Vit is proposed to solve the collar classification problem in complex scenes.This paper conducts multiple experiments on Collar6,CIFAR-10,and Deep Fashion6 to verify the superiority of MFA-Vit.All the models proposed in this paper have shown superior performance in multiple comparative experiments with existing methods,demonstrating that the models can be applied to collar classification tasks.The research in this paper can not only improve the accuracy of clothing quality evaluation but also provide a reference for the application of deep learning in the field of clothing design and production.