A power transmission line image generation method based on illumination feature attention mechanism

CN122156670APending Publication Date: 2026-06-05WUHAN UNIV OF TECH

View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: WUHAN UNIV OF TECH
Filing Date: 2026-02-13
Publication Date: 2026-06-05

AI Technical Summary

Technical Problem

Existing diffusion models suffer from loss of detail, brightness imbalance, and noise residue when generating images of transmission lines under low-light conditions, failing to effectively improve data diversity and recognition accuracy. Traditional attention mechanisms also fail to accurately capture the relationship between illumination distribution and structure in low-light scenarios.

Method used

An improved strategy based on illumination feature attention mechanism is adopted. Through a dual U-Net collaborative architecture and attention mechanism module, combined with variational autoencoder, U-Net denoising network and text encoder, feature extraction of low illumination images is optimized to generate micro-meteorological images of transmission lines that conform to the real low illumination patterns.

Benefits of technology

It significantly improves the brightness uniformity, detail clarity, and structural consistency of the generated images, provides high-quality and diverse training samples, and enhances the generalization ability and recognition accuracy of the micro-weather recognition model.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122156670A_ABST

Patent Text Reader

Abstract

The application discloses a power transmission line image generation method based on an illumination feature attention mechanism, and the method comprises the following steps: constructing a power transmission line weather image dataset and building an improved Stable Diffusion diffusion model, wherein the model comprises a variational autoencoder, a U-Net denoising network and a text encoder; inputting a low-illumination original image and a low-illumination enhanced text description into the model, encoding the low-illumination original image into a low-dimensional latent feature vector through the variational autoencoder, adding noise through forward diffusion as input, and meanwhile converting the low-illumination enhanced text description into a conditional semantic feature vector through the text encoder as the conditional guidance of the U-Net denoising network; generating a preliminary enhanced result and a semantic probability graph by using the U-Net denoising network; introducing an attention mechanism module, calculating an attention weight matrix, and performing secondary enhancement on the low-illumination original image to obtain a final power transmission line image. The application effectively improves the image generation quality.

Need to check novelty before this filing date? Find Prior Art