A train beacon image recognition attack method

Through the combination of feature attacks and multimodal coding models, the identification problem of train beacon recognition under complex visual interference and high-speed operation conditions is solved, and efficient and robust beacon information recognition is achieved.

CN119580228BActive Publication Date: 2025-05-09BEIJING AINIBABY HEALTH MANAGEMENT CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202510116002.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-01-24
Publication Date
2025-05-09
Estimated Expiration
2045-01-24

AI Technical Summary

Technical Problem

Existing train beacon identification and adversarial attack technology is difficult to deal with complex visual interference and accurate identification of beacon information during high-speed operation.

Method used

By determining the attack category, the image characteristics of the original train beacon image are extracted, and feature attacks are performed based on the attack category to obtain the simulated beacon image. Then, the simulated beacon image is input into the multimodal coding model, and the attention mechanism and the long and short-term memory network (LSTM) are combined to generate target image features and finally determine the train running status of the original train beacon image.

Benefits of technology

It realizes the robustness and effectiveness of accurately identifying train beacon information under complex visual interference and high-speed operation conditions, and improves the ability to recognize and attack the train beacon images.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119580228B_ABST
    Figure CN119580228B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of identification attack technology, and provides an identification attack method for train beacon images, the method comprising: determining the attack category of the original train beacon image, extracting the image features of the original train beacon image, performing feature attacks on the image features based on the attack category, and obtaining a simulated beacon image; inputting the simulated beacon image into a multimodal coding model to obtain target image features; the multimodal coding model includes an attention mechanism, and the key and value of the attention mechanism are a combination of the coding features of the simulated beacon image and a prompt sequence; inputting the target image features into a long short-term memory network to obtain the train running status of the original train beacon image. Different from the single attack interference mode in traditional research, a variety of attack methods can be selected, and modifications can be made to the image features; using LSTM combined with historical information to assist in judging the relationship between the current status and the overall status, enhancing the effectiveness of data recognition, and improving the robustness of the identification attack of the train beacon image.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Image description generation method and device and electronic equipment

    CN110717498A

  • Classification identification method based on convolutional neural network and multi-modal fusion

    CN115438708A

  • Detection method and system for resisting attack

    CN115880530A