Research On Clothing Style Recognition And Recommendation Based On Convolutional Neural Network

In recent years,with the rapid development of e-commerce platforms,more and more users choose online shopping.Online shopping has facilitated people’s life to a certain extent.However,due to the wide variety of products on e-commerce platforms,especially apparel products,which have strong personalization and rapid development,massive information appears in front of consumers,leading to the phenomenon of information overload.Although shopping platforms provide search functions for consumers,the number of queries is often very large.It is no doubt that consumers waste a lot of time and energy in choosing clothes one by one.In addition,clothing style is a relatively vague concept.Due to differences in living environment,education level and other factors,different people have different understandings of clothing style,resulting in deviation of style positioning.At present,there is no accurate definition and measurement of clothing styleFor this situation,this topic put forward a kind of fashion style based on convolution neural network identification and recommendation system,women’s clothing,for example,to identify the style on the basis of the recommended in style,and built a women’s clothing style recommendation system,combined with the user’s basic information and personal preferences to the user a satisfactory clothing style design is recommendedFirstly,through literature research and market survey,eight mainstream clothing styles on the market were determined,namely:classic style,elegant style,light style,casual style,sports style,neutral style,avant-garde style and national style.Through the analysis and arrangement of the eight styles,the characteristics of the eight styles were obtained.Secondly,8877 sample pictures of clothing styles were collected on TaoBao,T-mall,JD and other e-commerce platforms,and relevant experts in the clothing field were invited to screen the sample pictures,leaving 4410 representative sample pictures.Created a including classical style dress suit,classic style,classic style shirt trench coat,classic style,classic style,classic style skirts,elegant style dress,elegant style,coats,skirts,casual style elegant style dress dust coat,coat,leisure style,leisure style recreational style fleece jacket,recreational style,leisure style t-shirts,skirts,casual style recreational style Polo unlined upper garment sweater,leisure style,national style dress,coat,ethnic style,ethnic style jacket,shirt style,national style jacket,ethnic style skirts,avant-garde style suits,coats,avant-garde style Avant-garde style shirt,avant-garde style dust coat,avant-garde style dress,avant-garde style T-shirt,sports style dress,style movement style jacket,skirts,brisk style of dress,brisk style shirt,brisk style jacket,brisk style coat,coat,brisk style skirts,brisk style of neutral style suits,coats,neutral style style jacket 42 classes such as style of garment sample library.Thirdly,the convolutional neural network model of Alex Net was constructed,the weight parameters were adjusted through the back propagation algorithm,and the super-parameter files in the network model were modified to obtain the network model suitable for the recognition and classification of clothing style.The network model structure includes 5 convolutional layers,3 downsampling layers,3 full connection layers and 1 Softmax layer.The clothing style features are extracted through alternating convolutional and downsampling operationsFourth,data enhancement,grayscale and edge detection were used to preprocess the sample images,and the adjusted network model was used to train the preprocessed style samples for style recognition and classification.The recognition and classification accuracy was 97.45%.Fifthly,the sample pictures of identified and classified clothing styles are imported into the clothing style recommendation system as the systematic clothing style commodity model,and the clothing style commodity model is taken as part of the user preference model,and then the initial recommendation is made by combining the user’s style preference with the decision tree algorithm In order to dynamically capture users’ interest changes in time,the recommendation algorithm based on interest is used for dynamic recommendationFinally,the clothing style recommendation system was built by using Java language,MySQL database and other technologies,and the practical verification of the clothing style recommendation system was carried out.The method of mathematical statistics was mainly used to search for 50 subjects aged between 18 and 45 to log into the system for verification,and the subjects’ satisfaction with the recommendation result was counted.The results showed that the majority of subjects were satisfied with the recommended results,with an average satisfaction of 86.25.To sum up,this research focuses on the identification and classification of clothing styles,fine-tuning the Alex Net convolutional neural network hyperparameter file,and obtaining the method applicable to the identification and classification of clothing styles.Combined with Java language and MySQL database technology,an intelligent clothing style recommendation system was built,and the decision tree algorithm and interest algorithm were used to recommend the styles of users.