Research And Implementation Of Autonomous Development Of Garment Style Based On Deep Belief Network

In the field of garment, garment style is a very abstract category, it should be no doubt that it is very difficult to grasp. Garment style is a relatively complex process to identification and classification, and it is a concept with strong subjective, and complex decision factors, whose evaluation standard is strong personalized and style features is developing rapidly, all these features make the recognition and classification of garment style more complex. The traditional style quantitative and classification method could not meet the garment style’s characteristic of strong personalized and rapid development. Aim at the characteristic of garment style, we do research and implementation of autonomous development of garment style classification based on cognitive in these paper. The main achievements are as follows:Firstly, we raised a garment style classification method based on DBN (Deep Belief Network) autonomous development network which based on deep learning neural network. This method can resolve garment style’s several problems such as complex recognition and autonomous development, the insufficient to meet its features by traditional quantitative and classification method. It can adapt the garment style’s other features and solve the garment style’s personalized independent classification problems.Secondly, based on visual attention model of multi-objective MILN autonomous development cannot be a good solution to solve the large target line drawing, on this basis we propose a DBN model which based on the whole garment training, the goal of this model will be able to train a large number of line drawing, and also have certain to improve on accuracy, but the training time is much inferior to multi-objective MILN.Thirdly, the analysis method based on the whole style of garment cannot suit for the characteristic of training soon, we proposed an analysis method which based on the division feature of garment style, this method was based on the whole garment style feature autonomous development, and through the division feature style of line drawing of the network, then get corresponding to a particular style clothing in the garment style feature. And using the autonomous development of DBN solved the characteristics of whole garment training slowly and long cycling, and can effectively improve the training accuracy.At last, by integration of the above model and algorithm, we gives the analysis and comparison, which include the whole garment DBN and points feature DBN based on the garment style of autonomous development and the multi-objective based on MILN network.