| With the acceleration of urbanization in today’s society,the urban rail transit industry is developing rapidly,and the multi-network integration of intelligence and transportation has gradually become the development direction of future urban rail transit.Under the premise of ensuring safe driving,efficient transportation,energy saving and environmental protection,the society has put forward new requirements for the urban rail transit system,and the fully automatic operation system(FAO)of rail transit is playing an increasingly important role.Based on communication network,data collection and automation technology,the system highly integrates and deeply interconnects information such as vehicles,platforms,communications,and comprehensive monitoring.With the increasing complexity of the system operation,the resulting communication network problems,massive data processing problems,innetwork storage problems,unclear video images and other problems pose new challenges to the IP network-based FAO communication system.It has become a difficult point in the development of the FAO system.The Named Data Network(NDN)is different from the traditional network based on the TCP/IP protocol.It can not only realize efficient information forwarding and sharing based on data name identification,but also effectively improve the efficiency of data transmission and distribution in the network.It can also achieve effective guarantees for the mobility and security of network terminals.Therefore,this thesis introduces a data content-oriented communication method in the FAO system,which can fully meet the new needs of high-dynamic,high-efficiency,high-reliability,and high-speed mobile data exchange in future urban rail transit.Meanwhile,an image enhancement algorithm is proposed by combining Shearlet transform with deep learning to meet the FAO system requirements for the image quality.The main contributions of this thesis are as follows:(1)Based on the characteristics of FAO’s ad-hoc network,such as strong mobility,rapid topological structure change,and high real-time requirements,this thesis proposes an NDN-based FAO networking method,and uses the NS-3 network simulation platform to build and test.The communication performance of the FAO self-organizing network model.Experiments show that the use of data content-oriented communication methods can effectively improve the communication efficiency between FAO system nodes and fully adapt to the rapid changes in the network topology.The maximum endto-end delay of the communication process is 0.541452 μs,the transmission interference time in a high-speed mobile environment is about 290 ms,the user experience rate can reach 65765.67 Kbps,and the communication performance is excellent.(2)Based on the characteristics of large data volume and easy data congestion of FAO,this thesis proposes a machine learning-based NDN forwarding plane PIT table storage structure and data retrieval method LT-PIT to improve the data processing capacity of core routers and improve the purpose of network throughput.Experimental results show that under the condition that the false positive probability is less than 1%,the storage consumption of the index structure of LT-PIT is only 253.129 MB,which is much lower than the other schemes.(3)In order to meet the edge router requirements for low footprint in the NDNbased FAO communication subsystem,the memory efficient index called LCI-PIT for PIT and the data retrieval method are proposed.The experimental results show that the storage consumption of LCI-PIT is only 1.08 MB,and it can be easily deployed on SRAM to realize fast content name retrieval.The throughput can reach up to290.36 MSPS,which fully meets the system’s requirements for high throughput.Finally,Vivado HLS is used to implement FPGA implementation of the named data network forwarding plane workflow.The simulation results show that the FPGA implementation of the named data network forwarding plane workflow ultimately needs to use 724 lookup tables and 769 flip-flops,with a delay of 0.5 μs.(4)Aiming at the characteristics of FAO’s large scale of calculation,high requirements for image information,and complex image acquisition environment,this thesis first proposes a new type of preflight operator to improve calculation efficiency,and then proposes a method based on Shearlet transform and deep learning.The image enhancement algorithm.This method organically combines Shearlet transform with deep learning methods.First,the Shearlet transform is used to extract image features,and then the deep neural network is trained in the Shearlet transform domain.Then,in view of the instability of network training caused by the large statistical difference between the Shearlet transform coefficients,a dual-path training strategy and a data weighting algorithm are proposed.Finally,a deep convolutional neural network is built and the residual image to be obtained is learned from it.The simulation results show that compared with the original image enhancement algorithm,this algorithm can obtain higher quality images,which fully meets the requirements of the FAO system for image quality. |
