SAR图像显著性检测方法研究

发布时间：2018-01-09 18:04

本文关键词：SAR图像显著性检测方法研究　出处：《国防科学技术大学》2014年硕士论文　论文类型：学位论文

【摘要】：人类视觉系统具备高效的图像解译能力,能够快速地检测显著性区域,提取感兴趣目标。本文旨在将视觉注意机制理论应用于SAR图像解译中,提出适用于SAR图像的显著性检测方法。针对这一问题,本文总结分析了现有典型的显著性检测算法,并结合SAR图像的特性,提出基于显著性的SAR图像目标检测算法和尺度自适应的SAR图像显著性检测方法。本文主要工作包含以下几个方面:(1)总结分析了典型的显著性检测算法。首先重点分析了特征融合理论以及在此基础上提出来的Koch视觉注意生物神经学框架;讨论了显著性检测算法的客观评估指标;总结了四种典型的显著性检测算法的原理和计算方法,并使用上述算法分别对光学图像和SAR图像进行实验和评价。(2)提出了基于显著性的SAR图像目标检测算法。首先简要介绍了SAR图像背景杂波统计建模的方法,总结了SAR图像常用的背景杂波统计分布模型以及最优统计分布模型的选择准则。其次,总结分析了基于统计分布模型的双参数CFAR算法的算法流程和判决阈值的计算问题。再次,从视觉显著性理论出发,结合CFAR算法的窗口设计和SAR图像杂波背景统计建模方法,运用假设检验方法和贝叶斯定理设计基于显著性的SAR图像目标检测算法。最后,通过对比实验验证本文算法在虚警率、运算效率指标上优于基于统计分布模型的双参数CFAR算法。(3)提出了尺度自适应的SAR图像显著性检测方法。在Kadir显著区域检测算法基础上,结合SAR图像特性进行算法改进,提出了尺度自适应的SAR图像显著性检测方法。首先,重定义局部复杂度测度,解决信息熵度量方式不适合用于度量SAR图像局部复杂度测度这一问题。通过比率距离度量方式替代差值距离度量方式,克服SAR图像相干斑噪声,进而考虑了像素之间的空间分布,构造与空间分布相关的局部复杂度测度度量方法,该度量方法比信息熵度量方法更适用于SAR图像。其次,重定义了自差异性测度,选取了一种对于显著信息变化敏感的自差异性测度度量方法。再次,改进了显著性尺度确定方法,优化算法检测的准确性;最后,根据显著性测度和显著性尺度提出了显著图生成方法。实验结果验证了该算法比Kadir显著性检测算法更适用于SAR图像,并且,该算法的显著性检测效果优于四种典型的显著性检测算法。
[Abstract]:Human visual system has the ability of efficient image interpretation, can quickly detect significant regions and extract objects of interest. This paper aims to apply the theory of visual attention mechanism to the interpretation of SAR images. To solve this problem, this paper summarizes and analyzes the existing typical salience detection algorithms, and combines the characteristics of SAR images. This paper proposes a salience based SAR image target detection algorithm and a scale-adaptive SAR image salience detection method. The main work of this paper includes the following aspects: 1). The typical salience detection algorithms are summarized and analyzed. Firstly, the feature fusion theory and the Koch visual attention biological neural framework are analyzed. The objective evaluation index of salience detection algorithm is discussed. The principle and calculation method of four typical salience detection algorithms are summarized. The above algorithms are used to test and evaluate the optical image and SAR image, respectively. In this paper, a salience based target detection algorithm for SAR images is proposed. Firstly, the statistical modeling method of background clutter in SAR images is briefly introduced. The selection criteria of background clutter statistical distribution model and optimal statistical distribution model for SAR images are summarized. Secondly. The algorithm flow of two-parameter CFAR algorithm based on statistical distribution model and the calculation of decision threshold are summarized and analyzed. Thirdly, based on the visual significance theory. Combining the window design of CFAR algorithm and the statistical modeling method of SAR image clutter background, using hypothesis test method and Bayesian theorem to design a salient based SAR image target detection algorithm. Finally. Through the contrast experiment, the algorithm is verified in the false alarm rate. The operational efficiency is better than the two-parameter CFAR algorithm based on statistical distribution model. A scale-adaptive SAR image salience detection method is proposed, which is based on the Kadir salient region detection algorithm. Based on the algorithm improvement of SAR image characteristics, a scale-adaptive SAR image salience detection method is proposed. Firstly, the local complexity measure is redefined. The information entropy measurement is not suitable for measuring the local complexity of SAR images. Instead of the difference distance measurement, the ratio distance measure is used to overcome the speckle noise of SAR images. Then the spatial distribution of pixels is considered, and the local complexity measurement method related to spatial distribution is constructed, which is more suitable for SAR images than information entropy measurement. Redefine the measure of self-diversity, select a measure of self-diversity that is sensitive to the change of significant information. Thirdly, improve the method of determining significant scale and optimize the accuracy of algorithm detection. Finally, according to the significance measure and significance scale, a significant map generation method is proposed. Experimental results show that the algorithm is more suitable for SAR images than the Kadir saliency detection algorithm. The significance detection effect of this algorithm is better than that of four typical salience detection algorithms.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【参考文献】