Recognition Method for Train Coupler Handle Based on YOLOv5 Model

Abstract

To solve the problem of identifying different types of car couplers during the operation of the automatic uncoupling robot of a tippler, a method for recognizing the handle of a car coupler based on the YOLOv5 model has been proposed. This method selects YOLOv5n, which is relatively simple in the YOLOv5 series, as the benchmark model for the detection network. The overall structure is more concise, effectively reducing the number of model parameters while ensuring detection accuracy. The YOLOv5n model used for feature extraction and target recognition on two types of coupler datasets: upper action and lower action, greatly reducing the time required for training and testing, and achieving extremely high recognition accuracy. Compared with the commonly used SSD300 model and Faster R-CNN model, it shows significant advantages in terms of parameter quantity, computational complexity, predictive inference speed and weight file size.