基于深度学习的高铁电线杆及杆号检测与识别方法

Detection and Recognition Method of Pole and Number in High-Speed Railway Based on Deep Learning

面向高铁电线杆智能监测任务,提出一种基于YOLOv3的端到端的高铁电线杆自动检测及杆号识别算法。该算法首先对电线杆及杆号区域进行检测,并根据杆号区域检测坐标自动裁剪,然后识别杆号区域中的数字,最后将电线杆检测结果与数字识别结果自动结合。通过构建高铁电线杆图像数据集以及杆号区域数据集,进行大量实验。实验结果表明,我们提出的方法对电线杆及杆上编号的检测与识别准确率分别达到了97.50%、95.30%,能有效地完成高铁最优电线杆及杆号的自动检测任务。

This work offers an end-to-end automatic detection of poles and number identification technique in high-speed railway based on YOLOv3 for the intelligent monitoring task of poles in high-speed rail-way. To begin, the algorithm recognizes the pole and the number area, then crops automatically based on the detection findings of the number area’s coordinates. Then identify the numbers in the number area. Finally, the results of the pole detection and number identification are integrated automatically. By building the picture data set of poles and number region in high-speed railway, a great number of experiments were carried out. The experimental results show that the proposed method’s detection and recognition accuracy for the pole and the number on the pole is 97.50% and 95.30%, respectively, indicating that it can effectively complete the automatic detection tasks of the most optimal pole and the number on the pole in high-speed railways.