摘要
针对电力场景中的网状目标物提取难题,提出了一种基于超像素分割的电力故障识别算法。首先,在LAB空间上应用超像素分割算法进行分割,采用改进K聚类的方法生成网格簇;然后,针对网格簇分类困难的问题,提出了双重注意力机制MobileNet V2网络,分类后同类网格簇融合结果即为目标物掩膜;最后,在输电线路杆塔和换流阀巡检通道金属屏蔽网数据集上开展训练,获得了较高的准确率,并开展了边缘强化实验。
A recognition algorithm for power failures based on super-pixel segmentation is proposed as a solution to the challenging problem of extracting mesh targets in power scenarios.Firstly,the super-pixel segmentation algorithm is applied in LAB space to implement the segmentation,and improved K-clustering is used to generate mesh clusters.Secondly,for the difficulty to classify the mesh clusters,a dual attention mechanism Mobile Net V2 network is proposed,and a target object mask is extracted by the fusion result of similar grid clusters after the classification.The training is conducted on dataset comprising transmission line towers and metal shielding mesh for converter valve inspection channels,and the edge strengthening experiment is done,obtaining a higher accuracy and conducting.
作者
李渊
吴对平
杨瑞
包正红
曲全磊
沈洁
刘刚
LI Yuan;WU Duiping;YANG Rui;BAO Zhenghong;QU Quanlei;SHEN Jie;LIU Gang(State GridQinghai Electric Power Research Institute,Xining 810000,China;Key Laboratory of Safety and Defence of Power Transmission and Transformation Equipment of Hebei Province(North China Electric Power University),Baoding 071000,China)
出处
《智慧电力》
北大核心
2024年第12期43-50,共8页
Smart Power
基金
国家自然科学基金资助项目(U23B20135)
国网青海省电力公司电力科学研究院科技项目(SGOHDKYOSBJS2310236)。
