基于关键点匹配和深度特征的水电站无人机巡检图像异常目标识别  

Recognition of Abnormal Targets in UAV Inspection Images of Hydropower Stations Based on Key Point Matching and Depth Features

作　　者：崔杰连 于洋 CUI Jielian;YU Yang(State Grid Northeast Branch Luyuan Hydroelectric Power Company Taipingwan Power Plant,Dandong,Liaoning 118000,China)

机构地区：[1]国网东北分部绿源水力发电公司太平湾发电厂,辽宁丹东118000

出　　处：《计算技术与自动化》2025年第1期53-58,共6页Computing Technology and Automation

基　　金：国家电网公司科技项目(52991E20001U)。

摘　　要：为了满足不断上涨的电力能源需求,分布式能源并入大电网规模逐渐加大,尤其是水力发电机组(水电站),其占据比重最大,对电网运行安全性影响最大,需要保障水电站运行的可靠性,因此提出了基于关键点匹配和深度特征的水电站无人机巡检图像异常目标识别方法研究。预处理水电站无人机巡检图像,提升巡检图像的清晰度,以此为基础,采用方向梯度直方图模式提取巡检图像深度特征,选取较重要的深度特征作为关键点,对巡检图像进行匹配,将同一水电站场景巡检图像整合为一个集合,根据深度特征是否具有离群点判定巡检图像是否存在异常目标,计算异常目标深度特征与已知异常目标特征之间的相似度,从而实现巡检图像异常目标的识别。实验数据显示:应用提出方法获得的异常目标识别结果与实际异常目标识别结果一致,异常目标识别平均精度最大值为95%,充分证实了提出方法应用性能更优。In order to meet the constantly increasing demand for power energy,the scale of integrating distributed energy into the large power grid is gradually increasing,especially for hydropower units(hydropower stations),which account for the largest proportion and have the greatest impact on the safety of power grid operation.It is necessary to ensure the reliability of hydropower station operation.A method for identifying abnormal targets in unmanned aerial vehicle inspection images of hydropower stations based on key point matching and deep features is proposed.Preprocess the UAV patrol image of the hydropower station to improve the clarity of the patrol image.On this basis,use the directional gradient histogram mode to extract the depth features of the patrol image,select the more important depth features as the key points,match the patrol image,integrate the patrol image of the same hydropower station scene into a collection,and determine whether there are abnormal targets in the patrol image according to whether the depth features have outliers,Calculate the similarity between the depth features of abnormal targets and known abnormal target features,in order to achieve the recognition of abnormal targets in inspection images.The experimental data shows that the abnormal target recognition results obtained by applying the proposed method are consistent with the actual abnormal target recognition results.The maximum average accuracy of abnormal target recognition is 95%,fully confirming that the proposed method has better application performance.

关 键 词：异常目标识别 深度特征提取 水电站 关键点匹配算法 无人机巡检图像 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]