| In recent years,while the rapid development of China’s high-speed railway technology,railway safety issues have also received special attention,and the safety of contact network equipment plays a vital role.In order to further improve the efficiency and accuracy of contact network equipment detection,it is of far-reaching significance to realize the intelligent detection of video-oriented contact network equipment.Based on the contact network safety inspection device project in the 6C system,this paper proposes an image semantic segmentation neural network suitable for railway scenes(Deep Multi-Road Semantic Segmentation Network for Railway Scene,DMRN)performs semantic retrieval and localization on the key devices of the contact network,and uses the insulator as the object to detect the anomaly.Finally,the feasibility,universality and Strong adaptability of the image semantic segmentation algorithm proposed in this paper are verified by experiments in the complex railway environments.The main work and innovations of this paper are as follows:1.Aiming at the shortcomings of common segmentation algorithm in complex railway scenes,such as poor adaptability,weak universality and insufficient robustness,the multiscale receptive field and feature channel attention model are introduced,and an image semantic segmentation neural network(DMRN)suitable for railway scene is proposed.From the two perspectives of evaluation index and semantic segmentation map,the effect of six different sets of railway scene segmentation is evaluated.The average segmentation accuracy in clear weather reaches 91%,and the contact network under various complex weather conditions Segmentation of key equipment is still highly adaptable.Based on the semantic segmentation graph,the image retrieval technology is combined with the image processing technology to complete the semantic retrieval and positioning of the insulator,in order to prepare for its anomaly detection.2.Aiming at the problem that the insulator anomaly data is less and the fault type is unpredictable,the fault data simulation method of contact network critical equipment based on deep anti-neural network is studied,which enriches the insulator anomaly state data,and at the same time verifies the feasibility of the simulation method.The same fault data simulation method experiment was carried out.Therefore,the method studied in this paper provides a new idea to solve the problem of normal data and abnormal data imbalance of key equipment in the contact network in the railway scene.The improved CNN neural network is used to detect the anomaly of the insulator,and the correct rate is 92%.It is verified that the method has certain significance for the retrieval and location of the key equipment of the contact network and the anomaly detection.3.For the problem of tag data in semantic segmentation task is also image data,write pixel-level image tagging program,and manually complete the production of 300 tag images according to the annotation program,and simultaneously perform data on real image data and its corresponding semantic tag image data.Enhanced operations to provide sufficient data sets for the training of semantic segmentation neural networks. |
