Research on Key Technologies of Intelligent Detection of Substation Equipment Based on Cloud Edge Collaboration
Year: 2024
Type: conference-paper
Abstract: : The traditional manual inspection has some problems, such as strong subjectivity, time consuming and poor efficiency. With the remarkable progress of artificial intelligence, advanced communication and other technologies, and the wide application of deep learning models, the intensity of daily work has been reduced, but at the same time, new problems have been created. At present, the memory and computing power of the terminal hardware are limited, and when the model is deployed to the terminal side, the processing efficiency of the model is slow and the accuracy is low due to the limited hardware conditions. Aiming at the limited memory and computing power of the edge hardware, the lightweight network structure and model compression strategy are studied, and a channel structured pruning algorithm is proposed. A lightweight deep learning network model based on structured pruning of VGG-16 model was constructed. Experiments show that the optimized pruning model can obtain the same detection performance as the original model, and can reduce the size of the model by about 55.3% and increase the speed of 33FPS when the average accuracy mAP is lost by 8.1%, which provides effective support for the intelligent detection of substation equipment.
Authors: Ke Chang, Min Zheng, Chunpeng Wu
Institution: State Grid Corporation of China (China)
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4399257646