【摘要】 针对传统铁路接触网绝缘子识别方法准确率低、时间周期长和多尺度目标识别效果差等问题,提出一种改进Faster R-CNN的铁路接触网绝缘子识别算法.首先,对传统Faster R-CNN算法的特征提取网络进行改进,通过将主干网络中深层特征图像与浅层特征图像进行融合的方式获得语义信息度强、分辨率高的特征图像,解决多尺度识别的问题;然后,对传统非极大值抑制算法采用高斯降权函数进行优化,提高绝缘子识别召回率和准确率.仿真实验结果表明,改进Faster R-CNN算法对于绝缘子的识别精度达到了99.5%,召回率96.02%,较于原始Faster R-CNN算法分别上升了2.78%、3.04%.对铁路接触网绝缘子图像中不同尺度的目标可以有效识别,具有更高的准确率和召回率,为后续铁路接触网绝缘子故障检测提供了良好基础.更多还原

【Abstract】 Aiming at the problems of low accuracy of traditional railway catenary insulator identification methods, long time period and poor multi-scale target recognition effect, an improved Faster R-CNN railway catenary insulator identification algorithm is proposed.First, the feature extraction network of the traditional Faster R-CNN algorithm is improved, and the feature images with strong semantic information and high resolution is obtained by fusing deep feature images and shallow feature images in the backbone network to solve the problem of multi-scale recognition.Then, the traditional non-maximum suppression(NMS) algorithm is optimized by Gaussian weight reduction function to improve the recall rate and accuracy of insulator identification.The simulation results show that the improved Faster R-CNN algorithm has achieved 99.5% of insulator recognition accuracy and 96.02% recall rate, which haven been increased by 2.78% and 3.04% respectively compared with the original Faster R-CNN algorithm.The proposed method has a higher accuracy and recall rate, which can effectively identify targets of different scales in the image of railway catenary insulators and provide a good foundation for the subsequent fault detection of railway catenary insulators.更多还原