基于多尺度引导滤波和决策融合的电力设备热故障诊断方法研究

本文提出一种基于多尺度引导滤波和决策融合（multi-scale guided filter and decision fusion, MGDF）的电力设备热故障诊断方法，联合多尺度引导滤波和决策融合技术，充分挖掘红外图像的空间结构信息和温度信息。该方法有3个主要步骤。首先，基于热故障区域与环境背景在红外图像上的温度差异特性，逐像素计算热故障区域与环境背景的马氏距离，获取初始的热故障诊断结果。然后，采用不同参数设置的引导滤波器对初始诊断结果进行滤波处理，并将生成的若干引导滤波特征图堆叠在一起。不同参数下的滤波特征图包含着互补的空间结构信息。最后，为充分挖掘不同尺度特征图的空间结构信息和温度差异信息，利用主成分分析法对引导滤波特征图进行决策融合，提升热故障的诊断精度，生成最终的热故障诊断结果图。实验测试结果表明，本文方法在热故障诊断精度上有明显优势，满足电力设备红外巡检的应用需求。

Multi-scale Guided Filter and Decision Fusion for Thermal Fault Diagnosis of Power Equipment

This paper introduces a thermal fault diagnosis method called multi-scale guided filtering and decision fusion. The proposed method combines multiscale guided filtering and decision-fusion techniques for fault diagnosis. It comprises three main steps. First, the Mahalanobis distance between the fault area and background is estimated, and initial thermal fault diagnosis results are generated. The initial diagnosis result is then filtered using guided filtering with various parameters, and several filtering feature maps are generated. Different filtering feature maps contain complementary spatial-structure information. Finally, a principal component analysis algorithm fuses these filtering feature maps to capture their spatial structure information and thermal information in filtering feature maps. Experimental results show that the proposed diagnosis method has a better detection performance than the current state-of-the-art detectors.