Label data generation method and device and model training method and device

By generating straight lines, turnouts, and signal lights on the initial image, and performing section division and parameter adjustment, the problem of obtaining labeled data in the railway dispatching system is solved, achieving efficient and safe generation of labeled data and improving the model's recognition accuracy and generalization ability.

CN121120844APending Publication Date: 2025-12-12BEIJING URBAN CONSTR INTELLIGENT CONTROL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511295254.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-09-11
Publication Date
2025-12-12

AI Technical Summary

Technical Problem

In railway dispatching systems, existing technologies struggle to obtain high-quality labeled data, resulting in poor model generalization ability and security and privacy risks when acquiring labeled data.

Method used

By generating straight lines, switches, and traffic lights on the initial image, the system divides the track into sections, generates target images, and generates bounding boxes based on the positional information of the elements. By adjusting the track topology parameters and visual presentation parameters, the system generates a variety of labeled images, avoiding the need to obtain real-time images from the actual system for labeling.

Benefits of technology

It generated a large amount of high-quality labeled data, which improved the model's generalization ability, reduced labeling costs, avoided security and privacy risks, and enhanced recognition accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121120844A_ABST
    Figure CN121120844A_ABST
Patent Text Reader

Abstract

The invention discloses an annotation data generation method and device and a model training method and device, and relates to the field of computers.The annotation data generation method comprises the steps that a plurality of straight lines are generated on an initial image, a turnout is formed between every two adjacent straight lines through a connecting line, each straight line is subjected to section division, and a plurality of sections on each straight line are obtained; arranging signal lamps in a preset range of an end point of the distance section, generating a target image, generating an annotation box of each element on the target image based on the position information of the element, obtaining a basic image, performing one or more adjustment of a track topology parameter, a visual presentation parameter and annotation granularity on the basic image, and generating an annotation image set. According to the method and the device, when the annotation data is generated, a real-time picture is prevented from being acquired from a scheduling system in real operation and is annotated, and safety and privacy risks are avoided.
Need to check novelty before this filing date? Find Prior Art