Research On Classification Model Of Garment Recognition Based On Convolutional Neural Networks

With the rapid development of the Internet,e-commerce of clothing has emerged,such as Tmall and JD.People can easily buy the clothes they want with only a mobile phone.Image classification has gradually become an important research topic in computer vision,which has high research value and application value.Due to the large number of fine-grained classification and large visual changes,such as deformation,lighting,shooting Angle,background influence,and lens scaling,it is more and more difficult for the artificial design features to meet the realistic needs of image classification.The traditional method of clothing image classification is mainly based on the text description given by the image.However,with the increase of the amount of clothing image data,a large amount of manpower and resources are needed to manually label the images.Since everyone has a different perspective on the images,the labeling of the images will be different,which has a certain impact on the accuracy of the classification of clothing images.In order to improve the image classification performance,two methods of image classification,namely transfer learning and convolution network,are used to study the recognition and classification of clothing images.(1)The traditional clothing classification method is mainly to extract the color,texture,edge and other features of the image.These manual feature selection methods are tedious and have low classification accuracy.In order to improve the classification performance and timeliness of clothing images,this paper proposes a method of clothing image classification based on transfer learning based on convolutional neural network.The trained models(VGG16,Res Net50,Inception?v3,etc.)were transferred and trained on the garment image data set,the parameters of all the convolutional layers of the pre-trained model were retained,the network parameters of the front layer were frozen and the network models were fine adjusted to adapt to the recognition of the garment image.Experimental results show that the accuracy and timeliness of classification are improved effectively after the transfer learning.(2)Aiming at the problem of multi-category classification of clothing by the current clothing recognition and classification algorithm,the deep residual network can obtain a higher recognition accuracy by increasing the depth of the neural network,which is widely used in various fields.In order to improve the accuracy of garment image recognition,an improved depth residual network model is proposed in this paper.Introduce attentional mechanisms;Adjust the network convolution kernel structure.The network structure is tested on the standard data set Fashion-MNIST and the multi-category large clothing data set(Deep Fashion)provided by the multimedia laboratory of the Chinese university of Hong Kong.The experiment shows that the proposed network model is superior to the traditional deep residual network in the accuracy of clothing image recognition and classification.