【作者】 曾尚春；

【导师】 朱兆达；

【作者基本信息】 南京航空航天大学 ， 通信与信息系统， 2007， 博士

【摘要】 合成孔径雷达(Synthetic Aperture Radar,简称SAR)是一种有源微波成像传感器,它采用脉冲压缩技术获得距离向的高分辨率;采用合成孔径技术获得方位向的高分辨率。合成孔径雷达具有常规雷达无可比拟的高分辨能力,是现代雷达系统的重要发展方向之一。同时,波长较长的微波具有穿透植被和地表层的能力,是其他遥感设备,例如可见光和红外线等所无法比拟的。合成孔径雷达广泛应用于地形测绘、洪涝灾害监测、海洋污染监测、矿产森林资源和农作物普查、军事侦察等国民经济和国防领域。合成孔径雷达原始数据量很大,给信号的传输和存储带来很大困难,为解决这一矛盾,必须采用数据压缩技术。本文从算法性能和计算复杂度之间的折衷出发,对SAR原始数据压缩算法进行了较全面而又有重点的研究。在这一领域的一些常用算法基础上,研究了一些改进算法,并提出了一些新算法。文中给出了这些算法的SAR实测原始数据实验结果,并分析了各种算法的性能和计算复杂度。第一章阐述了合成孔径雷达原始数据压缩的作用和意义,介绍了这一领域的发展历程。分析了SAR原始数据统计特性,给出了常用的性能评价参数。研究了实测SAR原始数据的熵、概率分布以及方差的变化规律。分析表明去冗余压缩不适合于SAR原始数据,只能采用有损压缩的方法。本章最后给出了本论文的主要研究内容。第二章研究了SAR原始数据压缩的标量量化和矢量量化算法。分析了最佳标量量化器设计方法,给出了初始码本生成算法和三种码本设计算法。在块自适应量化(BAQ)算法的基础上,提出了一种改进的BAQ算法,在计算量相当的条件下提高了性能。根据SAR原始数据的幅度和相位特性,研究了幅度网格编码量化-相位均匀量化(ATCQ-PUQ)算法,该算法对幅度采用网格编码量化(TCQ),对相位采用均匀量化(UQ),该算法性能优于改进BAQ算法。针对三种不同原理设计的码本,分析了块自适应矢量量化(BAVQ)算法,矢量量化大大提高了编码性能。本章最后针对一些特殊数据块跟高斯分布偏离较大这一情况,提出了一种混合量化算法:块自适应标量-矢量量化(BASVQ)算法,该算法当数据块满足高斯分布时,采用标量量化;当数据块不满足高斯分布时,采用矢量量化。矢量量化时为减小计算量,采用了二叉树搜索算法,在计算量增加不多的情况下提高了性能。第三章研究了SAR原始数据压缩的变换编码算法。在分析了快速傅立叶变换-块自适应量化(FFT-BAQ)算法的基础上,提出一种改进的FFT-BAQ算法,使频域比特分配策略更加合理,在计算量相同的条件下提高了性能。研究了离散余弦变换-最佳熵约束块自适应量化(DCT-OECBAQ)算法,最佳熵约束算法能设计任意码率的标量码本,该算法性能优于改进BAQ算法。提出了小波变换-矢量量化(WT-VQ)算法,在分析了二级小波变换后的系数特性基础上,提出了二级小波变换后对平滑子带作无失真编码、对一级及二级细节子带作不同码率的矢量量化的编码策略,该算法性能优于DCT-OECBAQ,但计算量较大。第四章研究了距离聚焦后的SAR数据压缩算法。本章突破了常规SAR原始数据压缩的限制,采用先对SAR原始数据进行距离聚焦处理,以增加方位向的相关性,从而使SAR数据变得易于压缩。本章先分析了距离聚焦后SAR数据的统计特性,在此基础上提出一种距离聚焦-线性预测-块自适应量化(RF-LP-BAQ)算法,该算法先对SAR原始数据作距离聚焦处理,再沿方位作线性预测,最后对预测残差作块自适应量化。接下来提出了一种距离聚焦-沃尔什-哈达玛变换-块自适应量化(RF-WHT-BAQ)算法,该算法先对SAR原始数据作距离聚焦处理,再作二维可分离WHT变换,在变换域采用块自适应量化,量化器采用的是Lloyd-Max量化器。本章最后研究了距离聚焦-可变速率块自适应矢量量化(RF-VRBAVQ)算法,该算法充分利用了距离聚焦后的SAR数据方位向较强相关性这一特点,采用沿方位向的狭长分块,并沿方位向取量化矢量,运用合理的比特分配策略,采用多个码本作变速率矢量量化。RF-LP-BAQ算法和RF-WHT-BAQ算法性能都低于RF-VRBAVQ算法,但计算量大大减小。第五章结束语给出了本文算法得出的一些结论以及有待进一步研究和探讨的问题。更多还原

【Abstract】 Synthetic aperture radar (SAR) is an active microwave imaging sensor. It utilizes pulse compression technology to obtain high range resolution and aperture synthesis technology to achieve high azimuth resolution. SAR has a capability of obtaining high resolution that no ordinary radar can be comparable with, therefore, it is one of the significant directions for the modern radar development. At the same time, the microwave with long wavelength is able to penetrate foliage and the earth’s surface, which is unsurpassable by other remote sensing ways, such as visible light and infrared ray. SAR has been applied widely to many fields of civil economy and national defense, such as terrain mapping, waterlog surveillance, ocean pollution surveillance, forest and crop investigation, military reconnaissance, etc.The SAR raw data is difficult to be transmitted and stored due to large data size, so it is necessary to use data compression or reduction techniques. In this thesis, the compression algorithms for SAR raw data are studied comprehensively and emphatically from the trade-off between performance and complexity. Some existing algorithms are improved and some new algorithms are put forward. Experimental results of these algorithms using live SAR raw data are given, and the performance and complexity of each algorithm is analyzed simultaneously.Chapter 1 firstly elaborates the role and signification of SAR raw data compression, and outlines the history of its development. Then, the statistical characteristics of SAR raw data are analyzed, and performance evaluation parameters in common use are introduced. At the same time, SAR raw data’s entropy, probability distribution and the law of its variance are investigated respectively. Therefore, a conclusion is draw that the no-losing compression is not suitable for SAR raw data, and the losing compression has to be adopted. At the end of this chapter, the main contents and structure of this thesis are presented.In chapter 2, scalar quantization and vector quantization algorithms for SAR raw data compression are investigated. The structure of optimal scalar quantizer is analyzed. The algorithm of initial-codebook generation and three methods of codebook design are presented. On the basis of block adaptive quantization (BAQ) algorithm, an improved BAQ algorithm is put forward. Its performance is better on the condition of equal computational load. According to the amplitude-phase characteristics of SAR raw data, the amplitude trellis coded quantization-phase uniform quantization (ATCQ-PUQ) algorithm is investigated, in which the trellis coded quantization (TCQ) is applied to the amplitude, while the uniform quantization (UQ) is applied to the phase. It outperforms the improved BAQ algorithm. According to the three codebooks devised based on different principles, the block adaptive vector quantization (BAVQ) algorithm is analyzed. The encoding performance is improved much via vector quantization. At the end of this chapter, a combinational quantization algorithm, block adaptive scalar-vector quantization (BASVQ) algorithm is put forward. In this algorithm, the scalar quantization is adopted in the case that the data block satisfies Gaussian distribution. Otherwise the vector quantization is adopted. For the purpose of decreasing the computational load, bintree search is used in the course of vector quantization. The performance is improved with a little increase in computation.Transform coding algorithms for SAR raw data compression are studied in chapter 3. After analyzing the fast Fourier transform-block adaptive quantization (FFT-BAQ) algorithm, an improved FFT-BAQ algorithm is put forward. It makes bit allocation strategy in frequency domain more effective. Its performance is better on the condition of the same computational load. The discrete cosine transform-optimal entropy-constrained block adaptive quantization (DCT-OECBAQ) algorithm is investigated. The optimal entropy-constrained algorithm is able to design the scalar codebook with an arbitrary bit rate. Its performance is better than that of the improved BAQ algorithm. The wavelet transform-vector quantization (WT-VQ) algorithm is put forward. Furthermore, a coding strategy is proposed by analyzing the wavelet coefficients of the two-level wavelet transform. After two-level wavelet transform, no-losing coding is applied to the smooth subband and variable-rate vector quantization is applied to the one and two level detailed subband. Its performance is better than that of DCT-OECBAQ algorithm, but the computational load is rather high.In chapter 4, the compression algorithms for SAR raw data after range focusing are studied. The limitation of conventional SAR raw data compression is broken through by performing range focusing of SAR raw data firstly, which increases the azimuthal correlation and facilitates the compression for SAR data. In this chapter, the statistical characteristics of SAR raw data after range focusing are analyzed at first. According to that, a range focusing-linear prediction-block adaptive quantization (RF-LP-BAQ) algorithm is put forward. In the algorithm, the range focusing is performed for SAR raw data firstly, and the linear prediction is performed along the azimuth secondly, lastly, the block adaptive quantization is applied to the residual prediction series. In succession, a range focusing-Walsh-Hadamard transform-block adaptive quantization (RF-WHT-BAQ) algorithm is proposed. The algorithm fulfills the range focusing for SAR raw data firstly, and performs the two-dimension separable WHT secondly. At last, the block adaptive quantization is used in transform domain with Lloyd-Max quantizer. At the end of this chapter, range focusing-variable rate block adaptive vector quantization (RF-VRBAVQ) algorithm is put forward. It makes the best of the high correlation of the data in azimuth after range focusing. The long narrow block is adopted in azimuth, and the quantization vector is chosen along the azimuth. A reasonable bit allocation strategy is applied, and the variable-rate vector quantization is performed according to multiple codebooks. Although the performance of RF-LP-BAQ algorithm and RF-WHT-BAQ algorithm is not comparable with that of RF-VRBAVQ algorithm, their computational load is much less.Chapter 5 provides the conclusions of the algorithms proposed in this thesis, as well as the problems to be further studied.更多还原