Small sample image generation method for power defect scene

By building a diffusion model in the power defect scenario and introducing a texture modulation module, and combining the EMA algorithm to smoothly update the model parameters, the problem of low detection accuracy caused by data scarcity is solved, and high-quality and diverse defect images are generated under small sample conditions.

CN119693254APending Publication Date: 2025-03-25STATE GRID GANSU ELECTRIC POWER RESEARCH INSTITUTE +1
0 Cites 1 Cited by

Patent Information

Application Number
CN202411751167.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-02
Publication Date
2025-03-25

AI Technical Summary

Technical Problem

In the power defect scenario, data scarcity limits the training effect of deep learning models, resulting in unsatisfactory detection accuracy, and the existing GAN models lack generalization capabilities under small sample data conditions, making it difficult to generate high-quality and diverse images.

Method used

By constructing a diffusion model and introducing a texture modulation module, combining EMA algorithm to smoothly update the model parameters, the synthetic defective images are generated to supplement the actual samples, and the quality of image generation and the stability of the model are improved.

Benefits of technology

It realizes the generation of high-quality and diverse power defect images under small sample conditions, improves the stability and detection accuracy of the model, and alleviates the problem of data scarcity.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a small sample image generation method for a power defect scene, and relates to the technical field of image generation, and the method comprises the steps: S1, obtaining and preprocessing a power defect image, and building a training set and a test set according to the preprocessed image; s2, constructing a diffusion model, introducing a texture modulation module into the diffusion model, and training the diffusion model through the training set; s3, smoothly updating model parameters of the trained diffusion model by using an EMA algorithm; s4, inputting the images of the test set into the diffusion model after smooth updating of the model parameters so as to automatically generate small sample images of different power defect scenes; according to the method, the actual sample is supplemented by generating the synthesized defect image, so that the problem of data scarcity is relieved, and the quality of the generated image and the stability of the model are improved.
Need to check novelty before this filing date? Find Prior Art