基于改进SSD模型的高铁扣件定位算法

高铁扣件的检测对于保障铁路的正常运行起着十分重要的作用。针对高铁扣件定位精度不足以及传统定位算法无法定位道岔处扣件的问题,设计一种改进的SSD(single shot detector)深度学习扣件定位算法,即Improved_SSD。首先采用ResNet101更换经典SSD深度学习模型中的VGG16,增加网络深度的同时提高特征的抓取能力;然后利用膨胀卷积扩大网络的感受野,以不增加模型额外结构的方式提高模型的鲁棒性;最后提出一种非极大加权抑制方法,进一步提高扣件的定位精度。实验结果表明:与经典SSD算法相比,本文算法对扣件定位的召回率和精度分别提高了3.4%和4.7%;与其他几种扣件定位算法相比,本文算法不仅提高了对普通轨道扣件的定位精度,而且解决了传统定位算法无法定位道岔处扣件的问题。

High-speed Rail Fastener Positioning Algorithm Based on Improved SSD Model

The detection of high-speed rail fasteners plays an important role in ensuring the normal operation of the railway. Aiming at the problem of insufficient positioning accuracy of high-speed rail fasteners and the inability of traditional positioning algorithms to locate fasteners at turnouts, an improved SSD(Single Shot Detector) deep learning fastener positioning algorithm, namely Improved SSD, is designed. Firstly, ResNet101 is used to replace VGG16 in the classical SSD deep learning model, which increases the depth of the network and improves the grasping ability of the feature. Then, the expansion convolution is used to expand the receptive field of the network, and the robustness of the model is improved without increasing the extra structure of the model. Finally, a non-maximum weighted suppression method is proposed to further improve the positioning accuracy of the fastener. The experimental results show that, compared with the classical SSD algorithm, the recall rate and accuracy of the proposed algorithm are improved by 3.4% and 4.7% respectively. Compared with several other fastener positioning algorithms, the proposed algorithm not only improves the positioning accuracy of ordinary track fasteners, but also solves the problem that the traditional positioning algorithm cannot locate the fasteners at the turnout.