Research Of The Detection Algorithm On Railway Fastener Defects Based On Adaboost

With the development of the transportation industry,rail-haulage has become an indispensable part,which has greatly promoted the progression along the railway and made tremendous contribution to our economic construction with a diffuse and nationwide pattern.While ensuring the economic development,the technicians cannot ignore the challenge of security,especially the public transport part.The railway line is mainly composed of sleeper,rail,railway fastener and other parts.In order to avoiding the rail loose for a long time working,the railway fastener as a key connector is mainly applied to fix the rails with sleepers to effectively reduce the likelihood of accidents.The intact fastener plays a vital role to the normal operation of rail and sleeper,and the damaged or missing fastener may be the tremendous risk to the safety of train working.At the present stage,the detection of railway fastener is primarily manual inspection,but this way dependents on the multiple factors,such as the natural conditions,the energy and the technical levels of inspection personnel.However,with the speedy development of railway transportation in the number of trains,the speed of trains and the load,the traditional method of manual inspection can neither protect the personal safety of staffs,nor don't affect the punctuality of trains' running,so that this pattern cannot meet the modern requirements.Therefore,it is urgent to innovate an efficient and fast automated method to recognize the state of railway fastener.Through studying and analyzing the present situation of advanced technologies and mainstream literatures on detecting railway fastener,combinng the outstanding advantages of Adaboost algorithm,this thesis proposes an approach on railway fastener defects detection based on Adaboost algorithm.The main research contents are as follows:To efficiently reduce the amount of image information,the collection of sample images need to be pre-processed.Firstly,images 256 gray;Secondly,image enhancement by histogram equalization increases the comparison between the background and the fastener pieces;Thirdly,three filtering methods which are mean filtering,median filtering and the adaptive Wiener filtering are used to perform the simulation experiment.According to the denoising result,this thesis selects the adaptive Wiener filter to effectively reduce the noise interference,and decreases the complexity of the follow-up work.For the problems of previous fastener area locating,an innovative method is devised,which means adopting canny operator to detect the image edge,and regarding the result image as the input image;then using LSD line extraction algorithm to locate the lower edge of rail and the edge of sleeper;finally,basing on the prior knowledge to localize the final position of fastener.The outstanding innovation of this thesis is reasonably optimizing the problems ofAdaboost algorithm.At first,adopt the improved bacterial foraging algorithm to optimize the issue which exists in the searching strategy.Afterwards,present a new formula for calculating the weighting coefficient of weak classifiers;moreover,exhibit the detailed derivation.The weighting coefficient is not only relevant to the error rate,but also the capacity of recognition to positive samples and the reliability of weak classifier.At last,the effectiveness and efficiency of the optimization algorithm has been fully validated by the simulation of Heart data sets.The Haar-like features and the methodology to calculating the feature values are specified in this thesis.Furthermore,the integral image is used to compute the eigen values,and can effectively curtail the computation and enhance the operation speed.Then,select the appropriate Haar-like features of fasteners,and introduce a single rectangle.Fastener detection classification algorithm is designed based on the Adaboost algorithm,which includes training weak classifiers,devising strong classifiers and cascade classifier.The algorithm suggested in this thesis implements the recognition and classification of intact fastener,rupture fastener and missing fastener through the MATLAB simulation in Windows platform.The results of experiment show that the algorithm can achieve the automatic identification for the working status of the railway fasteners,and possess an essential theoretical guidance and economic practicality.