A REAL-TIME TRACK FASTENERS CLASSIFICATION SYSTEM BASED ON YOLOV4-TINY MODEL

Trains are one of the main transportations for people, and track maintenance has become an important topic today. In the past, maintainer walked through the track to see if the track is damaged or not, which is time-consuming and labor-intensive. This study utilizes Yolov4-tiny neural network to automatically identify 10 classes of track missing from the forward and 4 classes of track missing, 2 classes of normal track from the sideways in real time. Each of the precision is 92% and 96%, and the recall rate is 91% and 94%, which can effectively improve the maintenance efficiency and reduce the manpower demand.