计算机科学
人工智能
深度学习
云计算
错误检测和纠正
计算机视觉
感知
GSM演进的增强数据速率
功率(物理)
特征(语言学)
边缘计算
钥匙(锁)
实时计算
人机交互
作者
Lin Tian,Yuqing Zhou,Changyuan Guo,Dongdong Ma,Xiaotao He
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2025-09-27
卷期号:: 629-634
标识
DOI:10.1109/actce66599.2025.00124
摘要
This article focuses on deep learning (DL) based intelligent perception and behavior correction technology for power operation sites under cloud edge collaborative architecture, aiming to enhance safety monitoring capabilities in high-risk operation environments. Electric power operations have the characteristics of high risk and frequent accidents, and ensuring the safety of operators has become an urgent need in the industry. As the core task of intelligent security, human behavior recognition faces the challenge of processing massive edge video data. The traditional centralized cloud computing model has shown weakness in terms of real-time performance, bandwidth load, and response efficiency. Therefore, this article proposes a real-time anomaly behavior detection model that integrates cloud edge collaboration and DL. Firstly, design a keyframe localization algorithm based on keypoint motion features to accurately capture high-risk action segments at the job site; Secondly, innovatively reconstructing the spatiotemporal graph convolutional network (ST-GCN) with bottleneck residual modules enhances the ability to extract spatiotemporal features while maintaining model lightweight, thereby more efficiently identifying dangerous behaviors such as climbing and illegal operations. The experiment shows that the model has high recognition accuracy and low latency response capability in real power operation scenarios, and can achieve second level anomaly warning.
科研通智能强力驱动
Strongly Powered by AbleSci AI