Detection of rail faults from gray-level images with segmentation-based deep networks

Railway vehicles are widely used by people and load transportation in the world. In Turkey also with the high speed trains the railway transportation became an important part of our lives. These developments of railway vehicles are also causing the health of railways. It is because the number of vehicles, the cargo weights are increased and, with the new high-speed trains are on the rails. To obtain the maintenance of the rails, the health of rails must be checked in regular intervals, but it is hard due to the long kilometers of rails and with the traffic, it must be done at night. This cause negligence and mistakes to be overlooked. So, we need to do the controlling with an automatic computer system. In this project, focused on sleeper and rail problems, which are the two most important components of railway. Models are developed to do detection of cracks, defects in rail and checking the straight alignment of the sleepers by using U-NET deep learning algorithm. Although the trained models were used with insufficient and low-quality images, test results were obtained with an accuracy of 92% or more. Training models with sufficient quality and quantity of data is promising.