Substation reconstruction and extension violation behavior intelligent identification method based on meta-learning

Abstract

The invention discloses a transformer substation reconstruction and extension violation behavior intelligent identification method based on meta-learning, and the method comprises the steps: firstly, collecting a picture, constructing a difficult sample, completing the marking of a scene, forming a small sample data set, pre-training a YOLOv5 model on an ImageNet data set through employing a meta-learning method, and carrying out the fine adjustment on the collected small sample data set, and obtaining a final YOLOv5 model; secondly, deploying the trained YOLOv5 model to a mobile terminal, and completing the recognition of detection objects such as operating personnel, construction equipment, power transmission and transformation equipment and the like; and finally, setting a virtual electronic fence in a self-adaptive manner according to construction operation requirements, and carrying out intelligent recognition and alarm on boundary-crossing violation behaviors of personnel and machines based on the set virtual fence. The method is different from traditional physical fence and other types of virtual electronic fence technologies, not only can ground violation behaviors be effectively identified, but also high-altitude border-crossing violation behaviors can be identified, and the method is flexible in deployment, simple in operation, high in real-time performance and good in reusability.

Description

technical field [0001] The invention belongs to the technical field of intelligent identification, and in particular relates to an intelligent identification method for substation reconstruction and expansion violations based on meta-learning. Background technique [0002] The environment of the substation site is complex, and there are many factors that threaten people's lives. In the process of substation construction, it is necessary to quickly and accurately detect and give timely warnings to operators who are about to stray into dangerous areas, so as to reduce the loss of life and property. However, most of the existing dangerous area detection methods are based on sensors, which require a large number of sensors, and the actual operation is complicated. Moreover, attaching sensors to the workers will interfere with the construction operations of the workers, and manual observation is time-consuming and labor-intensive, which is easily affected by observers. mental st...

AI Technical Summary

Problems solved by technology

However, the existing technology mainly relies on a large amount of labeled data for model training, and then detects and recognizes at the back end, does not make full use of the front-end computing resources, consumes too much data transmission, and reduces applicability

[0003] The traditional video surveillance system is mainly manual monitoring, and video forensics is observed after the event; and the architecture design of the video surveillance system is relatively simple, it cannot perform real-time intelligent analysis of video images, cannot realize remote monitoring, and cannot provide timely alarms for abnormal events

Images