基于图像相似度与深度学习的变电站刀闸状态识别算法的对比研究

Comparative Study on Substation Isolating Switch State Recognition Algorithms Based on Image Similarity and Deep Learning

为解决以往变电站中基于各类传感器的刀闸状态检测方式成本高、稳定性差的问题,本文探索了两类基于图像识别的刀闸状态检测算法,相较于传统的基于图像相似度的刀闸状态识别算法,基于深度学习的目标检测算法对刀闸状态识别准确率更高,能够有效对变电站内刀闸状态进行检测。本文在对483张包含各类刀闸状态的图像进行标注后,使用YOLOv5的预训练模型进行训练,训练后的模型在包含80张各类刀闸状态的测试集上进行测试,结果表明综合准确率为89.31%,综合召回率为98.32%。本文所提出的基于深度学习的刀闸识别算法能够对变电站刀闸状态进行有效识别,且识别准确率高、部署较为简单,对保障变电站安全稳定运行有着重要作用。

To solve the problems of high cost and poor stability associated with traditional sensor-based circuit breaker status detection methods in substations,this paper explores two types of image-based circuit breaker status detection algorithms.Compared to traditional image similarity-based circuit breaker recognition algorithms,deep learning-based object detection algorithms achieve higher accuracy in circuit breaker status recognition and effectively detect the status of circuit breakers in substations.After annotating 483 images containing various circuit breaker statuses,this study utilizes a pre-trained model of YOLOv5 for training.The trained model is then tested on a test set containing 80 images of various circuit breaker statuses,yielding a comprehensive accuracy of 89.31%and a comprehensive recall rate of 98.32%.The proposed deep learning-based circuit breaker recognition algorithm can effectively identify circuit breaker statuses in substations,demonstrating high accuracy and simple deployment.It plays an important role in ensuring the safe and stable operation of substations.