Deep learning based fault detection of rail track components from images taken by autonomous drone

Rail transportation systems are a type of transportation that is widely used in passenger and freight transportation. Accidents in this type of transportation, which is widely used, cause serious loss of life and property. Rails, rail sleepers and fasteners are essential components for the train to move safely on the line while maintaining the symmetrical structure of the line. Errors such as breakage, deficiency or wear that may occur in these components should be detected and repaired. In order for the railway operation to be carried out regularly, maintenance must be done regularly. When maintenance is done manually by experts, mistakes can be high and take a long time. For this reason, rail defects can be detected using recently developed image processing algorithms and deep learning algorithms. In this thesis, image processing and deep learning-based new methods are proposed in order to detect errors in railway line components. The types of faults that may cause faults in the railway geometric structure have been examined with traditional image processing techniques and it has been determined that fault detection can be made at high success levels with the proposed methods. Deep learning-based new methods have been developed for the detection of errors in various components such as fasteners, sleepers, ballast, switch points in the railway, method steps and the results of the methods tested with real field images are given in the thesis. One of the most important parameters affecting the model success of image processing and deep learning algorithms is the quality of the processed data. Visuals obtained in accordance with the type of error to be detected increase the success of the model. The methods proposed within the scope of the thesis were carried out using images obtained with an autonomous drone. The advantageous aspects of obtaining images with an autonomous drone are that it does not require human labor, is independent of train services, and can collect images at different times of the day.