The periodic examination of wooden sleepers in railway tracks is examined with the help of human intervention. Using instance segmentation, the wooden texture on sleepers of railway tracks is detected, labelled, and masks are created for each class object. Segmentation is a combination of object detection, classification, and object localization. Mask R-CNN architecture is used to extract wooden sleeper regions from railway track images. The Mask R-CNN architecture is state of art in bounding box detection, keypoint detection and segmentation. Custom datasets are used for training the Mask R-CNN model. The custom dataset is prepared from drone, pre-processed and labelled using Make Sense AI tool. Then the model is evaluated based on IoU (intersection over union) of COCO dataset format.
Segmentation of overlapping ballast coverage on wooden railway sleepers using transfer learning technique
