基于改进机器视觉算法的电力设备红外检测

Infrared detection of power equipment based on improved machine vision algorithm

电力设备的运行情况较为复杂,所获取的电力设备红外图像易出现不完整的问题,导致设备的红外检测结果收敛性较差。为此,提出基于改进机器视觉算法的电力设备红外检测方法。构建改进机器视觉法的检测框架,通过卷积层进行端对端训练。从局部平移不变性角度出发,利用Mean shift聚类算法计算采样点概率密度。根据红外图像分割结果确定最佳目标图像尺度,以此为依据分析电力设备检测结果。采用多尺度小波变换的快速滤波技术完成图像处理,提取低频信号,并对其进行加权降噪,从而实现电力设备红外图像的增强。实验结果表明,该方法下红外图像检测结果完整,且训练损失值最小值为1,具有良好的收敛效果。

The operation of power equipment is complex,and the infrared image of power equipment is easy to be incomplete,resulting in poor convergence of infrared detection results.Therefore,an infrared detection method of power equipment based on improved machine vision algorithm is proposed.The detection framework of improved machine vision method is constructed,and end⁃to⁃end training is carried out through convolution layer.From the perspective of local translation invariance,the mean shift clustering algorithm is used to calculate the probability density of sampling points.According to the infrared image segmentation results,the best target image scale is determined,and the detection results of power equipment are analyzed.The fast filtering technology of multi-scale wavelet transform is used to process the image,extract the low⁃frequency signal,and weight it for noise reduction,so as to enhance the infrared image of power equipment.The experimental results show that the infrared image detection result under this method is complete,and the minimum training loss value is 1,which has a good convergence effect.