大场景PolSAR图像人造目标检测方法研究

发布时间：2018-12-27 12:16

【摘要】：人造目标检测是遥感数据应用的重要环节,是灾害救援、军事侦察等应用的基础,检测性能与速度的优劣直接影响到后续的实际应用。众多遥感数据中,PolSAR为主动成像,具有全天时全天候的特点,这在目标实时检测应用中具有极大的优势。目前PolSAR图像的获取能力获得了极大的提升,图像覆盖区域以及数据量越来越大,这为PolSAR图像目标检测应用提供了数据基础。然而,传统算法大多关注精度指标,当图像数据量过大时,检测处理耗时也随之大幅增加。而以人造目标检测为基础的诸多场合中,则需要在有限的时间内完成相关工作,这对大场景PolSAR图像人造目标检测提出了处理速度上的要求。如何在精度满足要求的条件下,高效完成大场景PolSAR图像的人造目标检测成为研究面临的主要问题。本文以大场景PolSAR图像人造目标检测为主要研究内容,首先从PolSAR图像表征形式与人造目标特性分析入手,明确了人造目标在PolSAR图像中的具体特征,在此基础上进行基本检测理论的研究。随后对传统的目标提取算法进行了加速优化,并设计了基于快速Wishart计算的检测算法。然后进行了图像均匀度描述因子特征的设计,该特征具有目标区域提取的能力,且提取速度与传统纹理特征相比获得了较大的提高。利用图像均匀度描述因子特征,设计了三种基于该特征的极化目标检测算法。根据图像信息使用量的不同,三种算法具有不同检测性能以及检测速度,以适应不同的精度以及效率需求。最后选择实验数据,分别进行陆地环境人造目标检测以及海洋舰船目标检测实验,并从目标检测精度、重点目标检测情况以及处理时间三个方面对算法进行评价。实验结果表明,本文设计的四种检测算法均具有大场景PolSAR图像人造目标检测能力。其中,基于快速Wishart计算的检测算法其性能与监督信息选取相关,具有较大的性能提升潜力。基于图像均匀度描述因子特征的三种检测算法,能够在较短的时间内完成人造目标检测任务,且检测精度比传统方法更高。三种检测算法对图像信息的使用量不同,其检测性能与处理速度具有较大的差异,能够适应不同的精度与效率要求。
[Abstract]:Artificial target detection is an important part of remote sensing data application, and is the basis of disaster rescue, military reconnaissance and other applications. The performance and speed of detection directly affect the subsequent practical applications. Among the many remote sensing data, PolSAR is active imaging, which has the characteristics of all-day, all-weather, which has a great advantage in the application of target real-time detection. At present, the ability of PolSAR image acquisition has been greatly improved, the area covered by the image and the amount of data are more and more large, which provides a data basis for the application of PolSAR image target detection. However, most of the traditional algorithms focus on the accuracy index, when the amount of image data is too large, the detection processing time is also greatly increased. In many occasions based on artificial target detection, it is necessary to complete the related work in a limited time, which puts forward the processing speed requirements for large scene PolSAR image artificial target detection. How to efficiently complete the artificial target detection of large scene PolSAR images under the condition of satisfying the requirement of precision has become the main problem in the research. In this paper, the detection of artificial targets in large scene PolSAR images is the main research content. Firstly, the characteristics of artificial targets in PolSAR images are defined by analyzing the representation form of PolSAR images and the characteristics of artificial targets. On this basis, the basic detection theory is studied. Then the traditional target extraction algorithm is optimized and the detection algorithm based on fast Wishart computation is designed. Then, the feature of image uniformity description factor is designed. The feature has the ability to extract the target region, and the speed of extraction is improved compared with the traditional texture feature. Three polarimetric target detection algorithms based on the feature of image uniformity description factor are designed. According to the different amount of image information, the three algorithms have different detection performance and detection speed to meet different accuracy and efficiency requirements. Finally, the experimental data are selected to carry out artificial target detection in terrestrial environment and marine ship target detection experiment, and the algorithm is evaluated from three aspects: target detection accuracy, key target detection situation and processing time. Experimental results show that the four detection algorithms designed in this paper have the ability to detect artificial targets in large scene PolSAR images. Among them, the performance of the detection algorithm based on fast Wishart computation is related to the selection of supervisory information, which has a great potential for performance improvement. Three detection algorithms based on the feature of image uniformity description factor can complete the task of artificial target detection in a relatively short time and the detection accuracy is higher than that of the traditional method. The three detection algorithms have different usage of image information, and their detection performance and processing speed are quite different, which can meet different requirements of accuracy and efficiency.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52

【参考文献】