Research And Implementation Of Fabric Image Retrieval Method Based On Improved AlexNet

Fabric is the raw material for all luggage manufacturing enterprises to produce products,and it is also the main material for the luggage material sales stall.Along with the advance of the Internet+and the need of improving the production efficiency of the enterprise,luggage manufacturer is eager to find the corresponding cloth material or similar material from the warehouse quickly and accurately through fabric information provided in product designing,in the same way,luggage materials sales archives mouths are anxious to get the corresponding cloth material or similar material from the warehouse according to the fabric color card information provided by the buyer.There-fore,it is of great application value to develop a system that can automatically retrieve the corresponding or similar material information from the inventory efficiently,quickly and accurately based on the color card image provided by the user.Traditional image retrieval methods have problems such as unsatisfactory retrieval accuracy when applied to luggage fabric image retrieval,especially when the retrieved fabric image and the sample library image have inconsistent image resolution or/and rotation angle which will lead to the poor retrieval effect.This thesis is devoted to solv-ing the aforementioned problems.Research on the model of fabric image retrieval has carried out based on deep learning,focusing on the improvement of the AlexNet net-work structure,and developed a fabric image retrieval system based on AlexNet opti-mization,which enable user to quickly and accurately retrieve the fabric color card information through the built model.The main work of the dissertation and the results obtained are as follows:(1)AlexNet network structure optimization.Aiming at the problems of tradi-tional text or content-based image retrieval methods,such as lack of learning ability,slow retrieval speed,low precision of image retrieval for rotation,multi-resolution,and complex backgrounds,a fabric image retrieval method based on convolutional neural network is proposed.In order to make this method more suitable for the field of fabric images,AlexNet network and Inception structure were combined to optimize the pa-rameters of the original AlexNet network structure and improve the network perfor-mance,making it more suitable for fabric image retrieval with different rotation angles and different resolutions.(2)Construction of fabric image retrieval model based on improved AlexNet.In order to solve the problems of time and memory consuming for large-scale data set retrieval,a fabric image retrieval model based on improved AlexNet model was de-signed,combined with the performance advantage of the improved AlexNet model in(1).First,the image to be retrieved is pre-classified to obtain its pre-classification iden-tifier Z,then all the feature vectors C that are the same as the pre-classification identi-fication Z are obtained from the feature database.Next matches similarity the feature vectors of the image to be retrieved with all the vectors in the feature vector set C,and finally the corresponding images and their types and numbers are returned according to the similarity.Experimental results show that this method can greatly reduce the retrieval time and memory consumption,and it is also very practical for large-scale image database retrieval.(3)Development of the system based on the improved AlexNet fabric image retrieval.According to the current demands of luggage enterprises on the image re-trieval of fabric image,the system of fabric image retrieval based on AlexNet was de-veloped.The system mainly consists of two modules:the client and the management.The client module consists of four sub-modules:registration and login,image retrieval,retrieval record query and personal center.The management end consists of five sub-modules:user management,fabric image management,fabric image storage,retrieval record management and model update.