Research On Baggage Retrieval Method Of Airport Warehouse Based On Improved CNN And Hash Code

With the increase in the number of civil aviation passengers and the amount of checked baggage year by year,there are more and more cases of abnormal baggage such as loss and damage.At present,the Civil Aviation Baggage Center still manages abnormal baggage manually.In order to help the Civil Aviation Department improve the efficiency of abnormal baggage management,this thesis has studied a great deal of literature and related technologies on image feature extraction,classification and retrieval.Aiming to realize fast and accurate baggage retrieval,this research also strives to design an intelligent warehouse baggage retrieval platform.The main contents are as follows:In terms of feature extraction and image classification,a baggage classification model based on an improved pre-trained deep convolutional network has been proposed for such problems as the different shooting angles of baggage images in the real environment and the high similarity of some types of baggage.First,this research improved the network structure and proposed a new convolutional combination structure AMB(Asymmetric and Moopling Convolution Block),which can extract features through multiple branches to enhance the expression ability of the network convolution layer,thereby improving the accuracy and robustness of image classification based on the network.Then it will use the L-Softmax metric learning loss for network training to enhance the discrimination of luggage features.Experimental results show that the newly-proposed baggage classification model has achieved a good classification effect,with an accuracy rate of 98.6% on the baggage test set.The features extracted by the model are also more distinguishable and the retrieval accuracy rate is higher.Experiments on the Cifar data set show that the recognition accuracy can increase by up to 1.2%after applying the AMB structure in different networks,which proved the effectiveness and versatility of the AMB convolutional combined structure.When it comes to image retrieval,in order to enrich the information contained in the features,a spatial pyramid pooling layer has been introduced into the improved baggage classification model,and a multi-layer feature fusion method based on the spatial pyramid pooling features and the fully connected layer features has been proposed as well.In addition,integrating with the mixed loss including classification error and quantization error,this research established a deep hash model that can be used for baggage retrieval.Experiments demonstrate that the proposed deep hash model has a better retrieval effect with its retrieval accuracy rate at 96.1%,and the single retrieval time and feature storage space are respectively1/6 and 2/5 of the original feature,which highly saved time and space.This research designed and implemented a warehouse baggage retrieval platform based on the proposed deep hash retrieval model.Including luggage image retrieval,luggage information management and other functions,this platform consists of two parts: a web management system and a We Chat applet application.Field tests and applications indicate that the proposed baggage retrieval model can finish abnormal baggage inquiry more accurately and swiftly.Besides,the developed baggage retrieval platform has comprehensive functions and stable performances,significantly reducing the workload of the personnel in Civil Aviation Baggage Center and realizing the intelligent management of abnormal baggage in the warehouse.