Research On Rail Fastener Center Position Method Based On Mask R-CNN Model

With the increasing speed of railway construction and railway transportation,it is becoming more and more important to build the intelligent railway technology system,to do well in the maintenance of line equipment and improve the maintenance level of the line.Rail fastener,as the middleware connecting rail and sleeper on the track,bears the responsibility of protecting the safety of railway trunk line,and is an important part of railway maintenance.At present,the installation and disassembly of rail fasteners are basically completed by manual work,which is not only inefficient,but also dangerous.The research content of this paper is to segment the image of the rail fastener on the spot through a deep learning model--Mask R-CNN,and then determine the central coordinates of each rail fastener in the image.The realization of this can prepare for the automatic loading and unloading of the rail fastener,which has certain engineering application value.The main work of this paper is as follows:(1)Based on the deep learning theory,this paper studies the working principle and basic structure of convolutional neural network(CNN),Faster R-CNN,FCN and Mask R-CNN.It focuses on the key technologies used in each model,such as RPN,Ro I Pooling,Ro I Align,FPN and Multi-task loss function.(2)Based on the principle of Mask R-CNN,this paper makes an in-depth study on the organization structure and code of the Mask R-CNN model,including the main functions and structures of each code file and module,especially the two core modules: Config module and Model module.(3)In this paper,the general case segmentation model Mask R-CNN is redeveloped,which is applied to the rail fastener center positioning project.Several contents related to the project implementation are mainly studied,including the establishment of the sample database,configuration and optimization of the model,training and prediction,etc.(4)In order to make the research subject of this paper more close to the engineering application and more intuitively feel the application effect of the Mask R-CNN model in the complex working environment,we developed the system on the basis of the open source Mask R-CNN model,and constructed a test system specially applied to the center positioning of rail fastener.This paper focuses on the design concept,main functions and system architecture of the system.At the same time,the hardware platform and software platform of the system are actually built on the workstation computer and deep learning server,and the necessary debugging and optimization of the system are carried out under the background of engineering application.(5)In order to verify the effectiveness of the second developed Mask R-CNN model and the self-designed rail fastener center positioning system,the test results are analyzed and studied.Based on the model,the working conditions of rail fastener under three kinds of complex background and interference conditions are further verified,and the test results under three kinds of interference conditions are analyzed and explained according to the false detection rate,missing detection rate,center positioning error and time consumption.At the same time,the performance comparison of the Mask R-CNN model after the second development and the original model under the same interference conditions.