摘要
针对传统异物入侵检测存在的问题,提出一种基于孪生神经网络的变电站异物入侵检测方法。用孪生神经网络进行运动前景提取,将背景图和待检图同时送入孪生神经网络,并且在后处理网络中引入注意力机制,最后通过连通域处理直接输出异物检测结果。此方法对异物没有预设检测类别的限制,另外设计的数据增强也进一步提高了算法抗抖动和抗光照变化能力。在测试数据集上取得的98.35%的准确率和98.61%的召回率证明了模型的有效性。
In view of the existing inadequacies of conventional foreign object intrusion detection,this paper proposes a substation foreign object intrusion detection method based on siamese neural network.The siamese neural network is used to extract the moving foreground.The background image and the image to be detected are input to the siamese neural network at the same time,and the attention mechanism is introduced into the post-processing network.Finally the connected domain processing is used to directly output the foreign object detection results.This method has no limitation on the preset detection category of foreign objects.In addition,the designed data enhancement also further improves the anti-jitter and anti-illumination change ability of the algorithm.An accuracy rate of 98.35%and a recall rate of 98.61%achieved on the test data set prove the effectiveness of the model.
作者
姬裕鹏
田鹏
柳杨
韩茂林
贾利伟
JI Yupeng;TIAN Peng;LIU Yang;HAN Maolin;JIA Liwei(CYG SUNRI Co.,Ltd.,Shenzhen 518052,China)
出处
《电工技术》
2024年第9期24-28,31,共6页
Electric Engineering
关键词
异物入侵
孪生神经网络
注意力机制
特征层融合
变电站
foreign object intrusion
siamese neural network
attention mechanism
feature layer fusion
substation
