【作者】 周正；

【导师】 严国萍； 柳健；

【作者基本信息】 华中科技大学 ， 信息与通信工程， 2007， 博士

【摘要】 近年来,随着卫星遥感技术的发展,高光谱图像的应用越来越广泛,例如,地球资源勘探、环境监测、军事侦察等。然而,随着图像光谱分辨率的不断提高,高光谱遥感图像的数据量也日益庞大,这给星上图像处理和卫星信道传输提出了诸多挑战。其一,宇航环境条件严苛,数据采集的代价昂贵,所以对图像的保真度要求极高,星上图像处理要尽量减少信息损失;其二,星载平台的处理能力有限,尤其是低负载、低功耗的设备决定了数据存储空间不可能很大,这就要求对采集的数据进行压缩,然后发送到地面及时处理;其三,无线信道的频谱资源极其有限,海量的遥感数据与有限的信道带宽间存在巨大的矛盾,因此需要高倍压缩待传数据,并尽量保证图像的质量;其四,卫星信道的不确定性较高,传输过程受到外界因素影响,所以对星载遥感数据的编码应考虑容错性。所有上述问题的核心在于如何综合考虑高光谱遥感图像的编码效率和编码质量,使其满足星载处理环境和各种应用的需求。本文以高光谱遥感图像压缩的理论和方法为主要内容,从编码的实时性和有效性入手,较深入地展开了两方面的研究:(1)高光谱图像压缩中的谱段预处理算法。为了充分挖掘图像的谱间相关性,提高高光谱序列光谱维编码的效率,研究其中的谱段重组算法,并结合谱间编码,研究适用于预测和整数变换的谱段预处理方法;(2)高光谱图像压缩中的空、谱变换算法。基于静态图像编码的相关理论,针对高光谱图像的特点研究适合星上高光谱数据处理的空、谱编码方案。具体的研究工作如下:从整体的角度出发,给出星载高光谱遥感图像软件压缩系统的框架结构和主要功能模块,对系统的工作流程与功能模块的处理机制进行了详细的描述。基于该框架,阐述其中主要模块的理论基础,包括预测和变换等图像编码算法与量化、熵编码等其他编码过程,并结合图像特征,分析这些理论用于高光谱数据压缩时的优点和局限性。研究高光谱图像压缩中的预处理方法。总结当前谱段预处理算法的研究现状,分析高光谱数据的空、谱特征;描述了C-MEAN聚类分组算法,并分析其局限性,探讨改进办法。以此为基础,提出了一种全新的基于自适应谱段分组的预处理算法。该算法充分考虑分组的依据,设计了符合实际应用的快速算法。通过对几幅高光谱实验数据的测试结果表明,修正后的C-MEAN算法的编码性能有所提高;而本文的自适应谱段分组算法能够在较低的计算复杂度下,大大提高高光谱图像的压缩性能,且不影响图像质量。研究高光谱图像压缩中的谱间去相关算法。在总结当前谱间编码算法研究现状的基础上,提出了一种精细的线性逐级分组算法,并基于该算法思想的灵活性,设计了可逆的光谱维整数DCT编码方法。精细的逐级分组从计算复杂度的角度考量,利用精细的非均匀间隔阈值组,对高光谱图像的谱段进行一次性分组。另外,基于自适应谱段分组算法,还提出了一种分段的线性最优预测方法。实验结果表明,逐级分组算法能较大地提高谱间去相关算法的性能,且实时性较好;基于逐级分组的整数变换方法与基于自适应谱段分组的分段预测方法均能进一步提高软件压缩系统的整体性能。研究高光谱图像压缩中的空间去相关算法。在总结当前静态图像编码算法研究现状的基础上,分析了高光谱数据空间域压缩时需要考虑的问题,并针对这些问题,设计了一种基于整数小波变换和位平面编码的压缩方法。然后,简单地分析了离散余弦变换比小波变换更适合高光谱图像的空间维编码,提出了一种基于二维整型局部变换和网格编码量化相结合的空间编码方法。实验结果表明,基于整数小波变换和位平面编码的空间域压缩方法进一步减小了计算复杂度,而结合整数DCT和网格编码量化的编码方法的性能也十分优越。建立高光谱图像的软件压缩系统。结合各主要功能模块的相关方法,设计了高光谱图像压缩系统,并为系统建立了异常情况处理模块,同时还构建了主、客观评价相结合的质量评价平台。实验结果表明,新的高光谱数据软件压缩系统体现出良好的性能。本文的研究工作覆盖了高光谱图像压缩系统框架的主要模块,具有理论研究意义和实际应用价值。更多还原

【Abstract】 In recent years, with the development of both satellite remote sense techniques, the applications based on hyper-spectral image become more and more popular, such as earth resource exploration, environmental monitoring and military reconnaissance. However, the amount of hyper-spectral data increases with the resolution of images. There are many challenges for image processing and channels transmission on satellite. Firstly, the environmental condition is harsh and the collection of image is generally expensive, so the information loss should be minimized during the image processing. Secondly, the processing capability of on-board platform is limited, particularly the low-power consuming devices have not enough storage space, and so the collection of data must be compressed before being transferred to the earth. Thirdly, there is a big contradictory between the limited communication capacity of satellite channel and large amount of hyper-spectral data, so it needs a tradeoff between data bit rate and the quality of image. Finally, satellite channels have high uncertainty. Therefore, the remote sensing data coding should have the ability of fault tolerance.All above factors require the cooperation of coding efficiency and coding quality so as to satisfy different needs of applications. Based on the framework of hyper-spectral image compression system, this paper aims at improving the efficiency and speed of real-time image coding by developing some new algorithms of these two techniques: (1) Band pre-processing algorithms for hyper-spectral image compression. To remove the spectral correlation and improve spectral coding efficiency, this paper studies the applicable spectrum regroup scheme for predictive coding and integer transform. (2) Spatial and spectral algorithms for hyper-spectral image compression. Based on the still image coding algorithms and the characteristics of hyper-spectral image, this paper investigates special and spectral encoding module, which is applied in processing of satellite hyper-spectral data. Specific research work is as follow.First, the basic theory of hyper-spectral coding is introduced and described. The framework and the main function module of the system are given. Based on this framework, the basal theory of the main modules is expatiated, including traditional prediction, transform algorithm, quantization and entropy coding. Second, research for band pre-processing of hyper-spectral image compression. Summing up the state of the art in band pre-processing algorithm and analyzing spatial and spectral characteristics of the hyper-spectral data, we propose an improved C-Mean clustering algorithm. On this basis, this paper proposes a new algorithm based on adaptive band regrouping and designs fast algorithm. The experimental results show that the proposal can maintain the quality of image under low computational complexity. So it is an efficient algorithm; the performance of the revised C-Mean coding algorithm has been improved.Third, research for spectral coding algorithms of hyper-spectral image compression. Based on the conclusion of the current spectral coding algorithm, this paper presents a detailed linear regroup algorithm. Based on the algorithm, the optimal linear prediction program and reversible spectral dimension integral DCT coding are designed. The experimental results show that these algorithms can greatly improve the performance of the spectral de-correlation algorithm.Fourth, research for spatial coding algorithms of hyper-spectral image compression. Based on a simple analysis of DCT and DWT, this paper proposes a new special coding program based on two-dimensional integer partial transform coding and trellis coded quantization. The experimental results show that the algorithm has superior performance by combining trellis coded quantization and DCT technique.Finally, the complete hyper-spectral image compression system is designed. This system has a module of handling abnormal. Considering the performance evaluation of the compression system, this paper designs a simple integrating platform with some subjective and objective quality criteria benchmark. The experimental results show that the new compression system has a good performance based on the evaluation platform. The research work involves the techniques in compression domains of hyper-spectral image; therefore, it has important theoretic and practical significances.更多还原