A rail transit data lightweight collection method based on deep clustering

CN122265792APending Publication Date: 2026-06-23CHINA RAILWAY SIYUAN SURVEY & DESIGN GRP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHINA RAILWAY SIYUAN SURVEY & DESIGN GRP CO LTD
Filing Date
2026-02-25
Publication Date
2026-06-23

AI Technical Summary

Technical Problem

In existing rail transit data processing technologies, lightweight acquisition of video image data suffers from high computational complexity and an inability to intelligently identify and filter high-value data. Traditional methods result in the loss of image details and high computational complexity, while deep clustering methods do not explicitly model the local manifold structure of the data, making them unsuitable for small- to medium-scale data scenarios.

Method used

We employ a deep clustering-based approach to extract low-dimensional features from rail transit video images using a convolutional autoencoder. This is combined with UMAP manifold dimensionality reduction and GMM Gaussian mixture model for intelligent data filtering, achieving efficient data processing and lightweight design.

Benefits of technology

It significantly reduces computational complexity, enables accurate identification and filtering of data structure, meaning, and value, reduces network load, is suitable for video and image data processing in multiple scenarios, and supports edge deployment and real-time response for intelligent operation and maintenance of rail transit.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122265792A_ABST
    Figure CN122265792A_ABST
Patent Text Reader

Abstract

The application provides a kind of track traffic data light weight acquisition method based on depth clustering, in view of the problems that the existing track traffic video image data light weight method has high computational complexity, cannot intelligently identify and screen high value data, the application realizes the low-dimensional feature extraction of high-dimensional video image data, local structure capture and intelligent clustering through the cooperative work of data preprocessing, convolution auto-encoder, UMAP manifold learning and GMM clustering.The application discards the additional training process of the traditional clustering network, significantly reduces the computational complexity, reduces the network load and improves the transmission efficiency, accurately identifies and screens the data structure, meaning and value, provides technical support for the edge deployment, real-time response and resource saving of track traffic intelligent operation, and is suitable for video image data light weight processing in multiple scenes such as maintenance, work detection, station and platform.
Need to check novelty before this filing date? Find Prior Art