Study On Size Recommending Of Clothing Methods Based On Back Propagation Neural Network

With the development of electronic commerce, apparel e-commerce has entered a rapid development period. With the increase in clothing sales network, apparel return rate has increased. Return rate is mainly due to the clothing that consumer purchase online do not fit. This is also the biggest obstacle to Internet apparel. The purpose of this research project is to find a fast, scientific, practical way to reduce the problem that clothing does not fit by online shopping.This paper chooses the Zhejiang area to the age of 18-25 years old young woman as a study, based on Fit sex and body size the close relationship. And the use of non-contact body measurement measured the size of data, total of 300 samples.First of all, based on the predecessors on the clothes fit-related research studies, studied the between of apparel fit characteristics and body parts characteristics, Established garment size selection based on the Analytic Hierarchy Process (The analytic hierarchy process, referred to as AHP ) . The model target layer is the most suitable garment size, criteria layer is the body's eight control sites, the program layer is Garments type (S, M, L, XL). By matching consumers of the control parts of the data and clothing to the data type number to obtain the final number to be consumers the right type of clothing. Effectiveness of the method is verified through the post, using the method of the subject matched the entire sample data.Secondly, because of artificial neural networks to solve nonlinear problems and the BP (back propagation) neural network model of the error back-propagation, designed the model of size recommended by BP network model: human control of parts as BP the network input, the corresponding garment numbers as the network output. The results of the AHP method as a sample of the training this network. However, BP neural network has inadequate and limitations, such as training a long time and converges to local minimum and so on, proposed three kinds of improved BP neural network and training the same sample data for the three kinds of neural network. The training speed and accuracy by comparison, Levenberg-Marquardt algorithm is determined the final model of this paper eventually and is applied in Garments size recommended. Finally, the weights of the trained models, the threshold to save, such a BP neural network can be used as an effective tool for other object to make the corresponding recommendation.In short, this article uses the different analytical methods and models to research and analysis fit the human body and clothing matching, then experiments show the feasibility and effectiveness of the method. These methods for garment size recommending.