基于高分辨率机器视觉的光伏电站智能巡检技术研究

Research on intelligent inspection technology of photovoltaic power station based on high resolution machine vision

为了提升光伏发电站的无人机巡检效率,针对无人机采集的高分辨率视频图像信息进行了识别算法研究。该算法采用一种残差块计算单元对传统卷积神经网络(CNN)的卷积运算加以改进,设计了恒等块以提升网络的非线性拟合能力,并通过卷积残差块降低网络中的参数数量。同时还引入了一种批量规范化运算,有效提升了网络的训练效率及健壮性。在某光伏电站的无人机巡检平台上进行的算法性能测试结果表明,引入残差计算单元后,所提算法的性能较CNN网络有了显著提升,对光伏变电站的故障识别精度提升了4.28%。

In order to improve the efficiency of UAV patrol in photovoltaic power stations,the recognition algorithm is designed for the high⁃resolution video image information collected by UAVs.The algorithm uses a residual block calculation unit to improve the convolution operation of the traditional Convolutional Neural Network(CNN),and designs identity blocks to improve the nonlinear fitting ability of the network,and reduces the number of parameters in the network through convolution residual blocks.At the same time,by introducing a batch normalization operation,the training efficiency and robustness of the network are effectively improved.The algorithm performance was verified on the UAV patrol platform of a photovoltaic power station.The algorithm performance test on the UAV patrol platform of a photovoltaic power station shows that after introducing the residual calculation unit,the performance of the algorithm has been significantly improved compared with CNN network,and the fault identification accuracy of photovoltaic substations has been improved by 4.28%.