Power violation behavior identification method, device and equipment and storage medium

CN121789266APending Publication Date: 2026-04-03ELECTRIC POWER RES INST OF STATE GRID ZHEJIANG ELECTRIC POWER COMAPNY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511160695.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-08-19
Publication Date
2026-04-03

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively identify complex violations in power operations, especially since traditional monitoring methods rely on inefficient manual inspections and target detection model-based approaches ignore the temporal and sequential nature of the behavior, resulting in low identification accuracy.

Method used

A multi-level feature fusion mechanism is adopted, which fuses skeletal heatmap features with RGB image features through a bidirectional cross-attention module, and extracts new skeletal heatmap features and RGB image features from the fused features through a self-attention module, and iterative processing is performed to improve recognition accuracy.

Benefits of technology

It improves the accuracy of identifying electricity violations, can more accurately capture the characteristics of electricity violations, and enhances the ability to identify complex action sequences.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121789266A_ABST
    Figure CN121789266A_ABST
Patent Text Reader

Abstract

The invention belongs to the field of power, and discloses a power violation behavior recognition method, device and equipment and a storage medium, and the method comprises the steps: extracting a plurality of skeleton node data from an RGB image, and generating a skeleton heat map; a first feature extracted from the RGB image and a second feature extracted from the skeleton heat map are input into the feature aggregation model, the cross attention module is used for fusing the first feature and the second feature to generate a fused feature, and the first self-attention module extracts a new first feature from the fused feature based on a lightweight multi-head self-attention mechanism; the second self-attention module extracts a new second feature from the fusion feature based on the lightweight multi-head self-attention mechanism, and the feature aggregation module generates an aggregation feature through channel-level connection of the first feature, the second feature and the fusion feature after multiple iterations; and inputting the aggregated features to a pooling layer and a full connection layer for classification to identify power violation behaviors existing in the RGB image.
Need to check novelty before this filing date? Find Prior Art