【作者】 赵振兵；

【导师】 苑津莎；

【作者基本信息】 华北电力大学（河北） ， 电力系统及其自动化， 2009， 博士

【摘要】 本文把图像配准引入到电气设备的双通道图像在线监测中,红外图像与可见光图像配准后可得到设备更丰富、更全面、更综合的互补信息,能提高电气设备实时在线监测的准确性与有效性,有利于保障电网的运行安全。本文首先构建了变电站电气设备双通道图像在线监测系统,研究了配准之前的预处理方法和红外与可见光图像的配准方法,主要工作如下:1.根据现场的实际情况,提出了一种变电站电气设备双通道图像在线实时监测的方案,构建了以红外传感器和可见光CCD(Charge Coupled Device,电荷耦合器件)传感器为核心器件的双通道监测的实际系统。系统实现了对设备的可见光图像和热状态信息的远程、实时、在线监测,提高了在线监测的可靠性和准确性。现场运行表明,该系统稳定可靠、效果良好。2.复杂的现场采集条件使获得的红外图像和可见光图像存在噪声干扰,配准之前必须对图像进行预处理。研究了两种消噪方法:①从估计图像中噪声的统计特性出发,提出了一种基于噪声标准差估计的小波图像去噪方法;②图像信号和噪声信号是由不同的物理源产生的,是相互统计独立的,将ICA(Independent Component Analysis)方法应用于图像去噪,提出了一种基于FastICA算法的去噪方法。实验结果表明两种方法都能提高图像质量,突出图像特征,为下一步红外图像与可见光图像的配准研究做了充分准备。3.通过理论分析与推导得出了基于经验模式分解的配准原理,并以互信息为配准测度,提出了一种基于剩余图像和互信息的红外与可见光图像的配准方法。大量红外与可见光图像配准结果验证了配准原理的正确性和方法的有效性及先进性,该方法取得了比传统互信息法、基于小波分解和互信息的方法更高的配准精度。对医学图像、遥感图像等多模态图像的配准也获得了较高精度的配准效果。4.把新颖且性能优秀的SURF(Speeded Up Robust Features)算法应用到红外与可见光图像的配准中,针对红外和可见光图像细节差别大甚至存在对比度反转和两者轮廓大体一致的特点,提出了一种基于边缘提取的SURF红外与可见光图像配准方法,解决了用红外负像配准的片面性问题和SURF算法对模态敏感的问题,从电气设备红外与可见光图像的配准结果可以看出该方法取得了非常快的配准速度和较好的配准效果,可满足在线实时的应用需求。更多还原

【Abstract】 Image registration is introduced into the dual-channel image on-line monitoring for power equipment, thus more abundant and comprehensive complementary information could be obtained through matching infrared and visible image, which could improve the accuracy and validity of real-time on-line monitoring for power equipment, and be benefit to guarantee the safe operation of power system. The dual-channel image on-line monitoring system for power equipment is established, and then the preprocessing method and the registration method of infrared/visible image of power equipment are studied in this paper. The main work is done as follows.1. The dual-channel image on-line monitoring method for power equipment in the substation is proposed according to the actual demand. A practical system with the infrared sensor and visible light CCD (Charge Coupled Device) sensor as the key devices is established. The system has realized remote, real-time and on-line monitoring of the power equipment’s visible image and thermal information, which could improve the reliability and accuracy. It is stable shown by its on-the-spot operation.2. The preprocessing of image before registration is inevitable because of the noise jamming in image acquisition under complicated field conditions. There are two de-noising methods studied in the paper.①Wavelet de-noising method is proposed based on the estimate of standard deviation according to the statistical characteristics of the noises.②The image and noise have mutually statistical independency, and ICA (Independent Component Analysis) is applied to image de-noising. So, the de-noising method based on the FastICA algorithm is proposed. Both methods could improve the quality of image and highlight image features, which make full preparation for the registration.3. The registration principle based on the Empirical Mode Decomposition (EMD) is obtained by theoretical analysis and deduction. On the basis of this principle, the infrared and visible image registration method is proposed with mutual information as registration measure. A lot of experimental results demonstrate that the registration principle is correct and the proposed method is advanced and effective. It has more precision compared with the traditional mutual information and wavelet composition method. The registration of typical multi-modal images such as medical image and remote sensing image has also high-precision results. 4. A novel detector and descriptor with high performance, coined SURF (Speeded Up Robust Features), has been introduced to the infrared and visible image registration. And the registration method is proposed based on edge extraction and SURF in view of the features that there is large difference, even contrast inversion and consistent contour between infrared and visible image in image details. The method can solve the problems of unilateralism of the infrared negative image being directly used to register and the modality-sensitivity of SURF. It has achieved good results and high speed according to the experimental results, and can fully satisfy the practical demand for real-time applications.更多还原