Diffusion model-based radar sensor data synthesis method and system

Through a generative adversarial network based on diffusion model, the problems of missing and noise in radar data are solved, and high-quality radar data is generated, which improves the environmental perception and safety of autonomous driving, intelligent traffic and drone navigation.

CN119181010AActive Publication Date: 2024-12-24UNIV OF SCI & TECH OF CHINA
View PDF 2 Cites 1 Cited by

Patent Information

Application Number
CN202411699537.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-26
Publication Date
2024-12-24
Estimated Expiration
2044-11-26

AI Technical Summary

Technical Problem

Acquisition of high-quality radar sensor data has difficulties in areas such as autonomous driving, intelligent transportation systems and drone navigation, especially in different weather and complex environments, resulting in data loss, incompleteness, noise and sparseness problems, increasing the number of problems. Cost and complexity.

Method used

Using a generative adversarial network based on diffusion model, linear embedding, position encoding and forward diffusion processing is performed through the combination of real pictures and radar data to generate high-quality synthetic radar data, and optimize the generation model through discriminator to solve data loss and noise question.

Benefits of technology

It improves the authenticity and diversity of radar data, improves the system's environmental perception ability and operational safety, has high computing efficiency and generation stability, and is suitable for a variety of complex environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119181010A_ABST
    Figure CN119181010A_ABST
Patent Text Reader

Abstract

The invention discloses a radar sensor data synthesis method and system based on a diffusion model, and relates to the technical field of radar signal processing, and the method comprises the steps: inputting a real picture into a generative adversarial network, so as to output high-quality synthetic radar data; the training of the generative adversarial network comprises the following steps: acquiring a real picture and real radar data corresponding to the real picture; processing a real picture by using an encoder to obtain a semantic vector; performing forward diffusion processing on real radar data, and recording state data when the standard Gaussian noise is added in each step; inputting the semantic vector and the state data of each step into a decoder of a generator to generate synthetic radar data; inputting the real radar data and the synthetic radar data into a discriminator to generate a real radar feature vector and a predicted radar feature vector; according to the radar sensor data synthesis method and system, vivid radar data can be generated in various complex environments, and reliable data support is provided for an intelligent system.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Radar individual identification adversarial sample generation method based on generative adversarial network

    CN117648574A

  • Medical image segmentation method based on u-shaped network

    WO2022199143A1

Cited By

  • Lightning wave generation method and system based on diffusion model

    CN120386986A