Track obstacle zero sample classification method and device for air-based monitoring platform

Abstract

The invention discloses a track obstacle zero sample classification method and device for an air-based monitoring platform, and relates to the technical field of aviation monitoring and track obstacledetection. The device comprises a picture collector, an obstacle target extractor, a target mapping network, a semantic vector generator, a semantic vector mapping network, a nearest neighbor classifier based on cosine measurement and an alarm. The method comprises the following steps: acquiring a rail picture through an unmanned aerial vehicle, extracting an obstacle target area, and mapping theobstacle target area into a fixed-dimension vector as a visual feature; generating a semantic vector for the category name of the obstacle by using a Word2vec technology, and mapping the semantic vector into a semantic feature vector with the same dimension as the visual feature; establishing a nearest neighbor classifier based on cosine measurement to classify obstacles; and in the training stage, training the mapping network by using the visible class obstacle samples. According to the invention, the detection of unknown obstacles on the rail is realized, the classification accuracy of therail obstacles is improved, and the false detection and false alarm rate is reduced.

Description

technical field [0001] The invention belongs to the technical field of aviation surveillance, and in particular relates to a zero-sample classification method and device for orbital obstacles of a space-based surveillance platform. Background technique [0002] The space-based surveillance platform is used to ensure the normal operation of the rail transit system on a large scale throughout the day. One of the important tasks of the platform to maintain the rail transit system is to accurately detect rail obstacles. [0003] Traditional track obstacle detection is divided into real-time detection and non-real-time detection. Non-real-time detection mainly includes detection vehicle troubleshooting and manual inspection. These two methods are to carry out troubleshooting before the equipment is in operation to ensure that the track can be used normally. However, if the track failure cannot be found in time between the two investigations, it is easy to cause serious vehicle...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that the current space-based real-time detection system is difficult to detect obstacles due to the lack or lack of training samples when detecting orbital obstacles, the present invention proposes a zero-sample classification method and device for orbital obstacles on a space-based monitoring platform. It can realize the orbital obstacle inspection of the space-based surveillance platform in the extreme situation where the obstacle target category is invisible (no training samples), and improves the classification accuracy of orbital obstacles, and reduces the false detection and false alarm rate

Images