R-YOLO orbital personnel target detection model

Abstract

Aiming at the problems of low accuracy and poor real-time recognition of railway personnel intrusion, an R-YOLO track personnel target detection model was proposed based on the YOLOv4 model. Firstly, the original CSPDarknet53 network was replaced by the lightweight backbone network ResNet50, and the standard convolution in PANet was replaced by deep separable convolution, which reduces the number of network layers and model volume, and accelerates the model recognition speed. Secondly, the effective channel attention module was added to the three feature layers before and after strengthening the feature extraction network, and the K-means++ clustering algorithm was used to re-cluster and analyze the dataset to improve the accuracy of the target detection model. In terms of model training, transfer learning and hybrid dataset joint training were used to solve the problems of poor personnel identification accuracy,false detection and leakage detection. Finally, the R-YOLO track personnel target detection model was used to test the real railway personnel intrusion dataset. The experimental results show that the average recognition accuracy of the R-YOLO model on the real railway personnel intrusion dataset reaches 92 .12%, which is 1 .89% higher than that of the traditional YOLOv4 algorithm, and the frame rate increases from 38.74 f·s-1 to 47.73 f·s-1. The R-YOLO model improves the problem of false detection and leakage of railway intrusion personnel, improves the real-time and accuracy of railway personnel intrusion identification, and provides guarantee for the safe operation of railway.

Keywords

• computer image processing; track detection; personnel identification; r-yolo; cnn; lightweight