Clothing Matching Intelligent Recommendation System Based On Deep Learning

With the continuous improvement of life,wearing warmth is far from meeting the requirements of the public.How to wear the right mix and how to wear it in different scenes has become the goal pursued by the public.With the development of the modern clothing industry,the huge number of clothing products makes it difficult for consumers to choose the right mix,and the traditional methods of matching are mostly recommended by online experts or recommended by the clothing matching library.Due to the particularity of the clothing image,clothing styles and styles are difficult to consider.The development of artificial intelligence provides a good idea and method for clothing matching.Convolutional neural network can better extract the style of clothing and more features that are not obviously noticeable,and can make better recommendations.As a new industry,clothing smart collocation has great research significance.Compared with traditional clothing collocation,this paper proposes a clothing matching method based on Siamese and deep hash for the diversity and complexity of clothing styles and styles.The roll machine network processes the clothing image,and the clothing image is vectorized and matched.According to the extracted semantic features of the clothing image,the recommendation is given.The paper mainly identifies and segmentes the clothing image,and extracts the clothing feature.The top and bottom of the clothing was studied.First,a Mask-RCNN network is trained on the Deep Fashion2 training set to identify and segment the garments.Then the processed top-loading images are imported into the Siamese network,and the features of the garments are extracted using Res Net.Style and costume posture;afterwards,the extracted features are vectorized,processed by the hash layer to generate the hash code of the garment;finally,by calculating the Hamming distance of the upper and lower loading,the method of query expansion is used to complete the matching of the garment.The model adopts a training method based on the Siamese network structure,so that the hash code retains the semantic information of the clothing image as much as possible.Based on the above model,the self-built street-show data set is used for verification.Experiments show that the method has a better hash rate when the hash length is 16,and the clothing matching is improved.The accuracy is suitable for everyday clothing mix.