Applied Research Of Deep Learning In LKJ Basic Data Management System

LKJ basic data is the fundament of train operation.It is crucial for the safety of train transportation to standardize the management of basic data strictly.Existing LKJ basic data management methods have weak points,like numerous management links,large amount of repetitive work,and high probability of errors.Therefore,it is necessary to explore scientific and effective basic data management methods to improve work efficiency and reduce the probability of errors.In this thesis,two problems in LKJs basic data management are solved using deep learning algorithms: First,user identification in the process of data approval and circulation;Second,automatic verification between data generated by different departments.Signature identification is a widely used identity authentication method.The signature identification based on the Siamese neural network has been proved to be feasible.This thesis will improve the accuracy and robustness of the signature identification through the improved Siamese network.At the same time,by identifying the railway yard schematic diagram of the electricity service,the data verification between the electric affairs department,the public works department,the maintenance department and other departments is realized,and the workload of manual calibration is reduced.The main results of this thesis are as follows:(1)A Sig UNet network with dual attention mechanism is proposed for signature authentication.The U-shaped structure of UNet is used to fuse multi-scale features,which improves the network feature expression ability.The Transformer self-attention mechanism and SENet channel attention mechanism are added,and the global semantic information is used to guide the extraction and screening of features,which improves the effect of the model.The experimental results show that the accuracy of this method on Bengali,Hindi and other public datasets is improved by about 10% compared with the benchmark method,being at the leading level compared with advanced algorithms,and it has high accuracy and robustness for small sample problems.(2)Combined with the characteristics of railway yard schematic diagram of the electricity service,the identification algorithm of the railway yard schematic diagram is designed,and the obtained information is used to verify the basic data.Using multiple methods like template matching algorithm and corrosion expansion algorithm,the location of various devices in the railway yard schematic diagram can be detected in a simple and direct way.On this basis,two straight line detection algorithms,Hough transform and LSD,are used to detect straight lines,and then identify the topological connection between devices.DBNet is used to detect the text area of the text number of each device,and after character-level segmentation processing,Le Net is used to identify the specific text content.Finally,a depth-first search algorithm is used to traverse the switch nodes,which can verify whether the position of the kilometer mark in the data table corresponds to the topological connection relationship.The experimental results show that the method in this thesis can effectively identify the information and topology connection relationship of each device in the railway yard schematic diagram,and can effectively check the data of five departments including the public works department,the electric affairs department,and the transportation department,which greatly reduces the workload of manual checking of the LKJ basic data.