Method for detecting railway foreign matter invasion in special scene based on transfer learning

Abstract

The invention provides a method for detecting railway foreign matter invasion in a special scene based on transfer learning. The method comprises the following steps: S1, collecting data; S2, data expansion: converting the image into images of different domains by using an image style converter to realize migration of an image style; S3, data processing; S4, target detection and tracking. Firstly,aiming at the problem that training samples are too few in severe weather environments and night scenes, conventional scene samples are migrated to corresponding scenes, and a railway scene target sample library is established under different weather and time conditions; and secondly, based on a deep convolutional neural network, through combination of a target detection SSD algorithm and a CAFFEframework, detection of an abnormal target in a railway scene is realized, and through combination with the generated complex sample library, model training under complex weather conditions based ontransfer learning is realized, and target detection, tracking and behavior analysis of the model are utilized.

Description

technical field [0001] The invention relates to the field of railway foreign object intrusion detection, in particular to a method for detecting railway foreign object intrusion in a special scene based on transfer learning. Background technique [0002] The entry of high-risk foreign objects such as pedestrians, animals, and falling rocks into the railway perimeter will pose a huge threat to the safety of trains. Therefore, real-time detection and early warning of abnormal events is of great significance to ensure the safety of train operation. At present, there are mainly two methods for the detection of railway foreign object intrusion: contact railway foreign object detection and non-contact railway foreign object detection. Contact detection technology includes dual grid deployment detection method, fiber Bragg grating detection, etc.; non-contact detection technology includes radar technology, infrared radiation technology, video-based analysis, etc. Among them, the ...

AI Technical Summary

Problems solved by technology

[0005] The training samples used in existing research methods are too simple to achieve good results for complex image detection of special weather, night scenes and railway scenes, so it is difficult to meet the requirements in practical applications, and the use value is low

In order to meet the real-time detection effect requirements of the specific scene, it is usually necessary to manually collect a large number of learning samples, but the sample collection of special weather and night scenes in the railway scene is difficult, the sample size is small, and it is difficult to achieve a good learning effect, and it is difficult to carry out the detection effect. promote

Images