Method and apparatus for recognizing inspection image of power system, device, and storage medium

By combining multimodal pre-trained models and bias loss calculation, the overfitting problem of power system inspection image recognition models in the case of few samples is solved, and accurate recognition and analysis of power system regional inspection images are realized.

WO2026137894A1PCT designated stage Publication Date: 2026-07-02GUANGDONG ELECTRIC POWER SCI RES INST ENERGY TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
GUANGDONG ELECTRIC POWER SCI RES INST ENERGY TECH CO LTD
Filing Date
2025-08-13
Publication Date
2026-07-02

AI Technical Summary

Technical Problem

Existing power system inspection image recognition models require a large amount of labeled data during training, making it impossible to directly and effectively recognize inspection images. Furthermore, they are prone to overfitting when there are few samples, resulting in insufficient generalization.

Method used

A multimodal pre-trained model structure is adopted. The image category recognition model is trained using a sample set of regional inspection images of the power system. The similarity between image features and category prototypes is calculated. Bias loss is introduced to avoid overfitting, and an initial image category recognition model is constructed.

Benefits of technology

It improves the recognition accuracy and generalization ability of power system area inspection images, enabling accurate identification of equipment status with few samples and avoiding the problem of model overfitting.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2025114516_02072026_PF_FP_ABST
    Figure CN2025114516_02072026_PF_FP_ABST
Patent Text Reader

Abstract

Disclosed are a method and apparatus for recognizing an inspection image of a power system, a device, and a storage medium. The method comprises: acquiring an inspection image to be recognized of a power system region; and inputting the inspection image into an image category recognition model to extract and obtain a first image feature, determining, on the basis of category prototypes in the image category recognition model, first similarity levels between the first image feature and the category prototypes, and determining, on the basis of the first similarity levels, a corresponding category prototype as a device state category of the inspection image, the image category recognition model being obtained by means of training an initial image category recognition model using an inspection image sample set of the power system region. The present invention solves the technical problems in the existing technology of low accuracy of recognition of an inspection image of a power system region, the inability to directly recognize and analyze the inspection image, and the inability to effectively recognize the inspection image of the power system region.
Need to check novelty before this filing date? Find Prior Art