【作者】 李耀辉；

【导师】 吴冬梅；

【作者基本信息】 西安科技大学 ， 信号与信息处理， 2009， 硕士

【摘要】 超光谱图像在军事和民用领域中的目标分类、目标识别、目标跟踪等方面具有重要的研究价值和应用意义。随着光谱成像技术的发展,光谱成像仪分辨率的提高,超光谱图像作为一种重要的数据源,其数据量日益庞大,给存储和传输都带来了巨大的压力。而到目前为止,一直没有形成一套成熟或标准的超光谱图像压缩技术。因此,对超光谱图像压缩编码的研究具有重要的应用价值。由于超光谱图像具有细节丰富,容量大的特点,采用传统的压缩方法都存在不同的局限性,而要实现超光谱图像这样的大数据量的快速无损压缩就更难了,目前以实现近无损压缩的研究居多。本文选用以推扫形式获得的超光谱序列图作为研究对象,依据特殊的图像获取手段带来的独特的图像特征,建立了简单DPCM谱间预测模型。在处理预测后得到的残差图像时选用LeGall5/3整数小波基对其作基于提升的整数小波变换。该运算只涉及加减法和移位运算,且可以获得完全还原的小波变换系数,能满足易于硬件实现和无损压缩的要求。本文用C语言编写了超光谱图像数据读取模块、谱间预测模块、整数小波变换模块、嵌入式零树编码模块的相关程序,并利用CCS开发环境和闻亭公司的TDS642EVM开发板对所设计编写的算法进行了成功的实测。最后结合CCS集成开发环境和TDS642EVM开发板,采取编译器优化、软件流水线技术、数据打包处理、DMA技术搬移数据、改进配置和存储空间等方法对算法进行了优化,大大提高了硬件资源的利用率和系统的实用性。实验表明使用本系统对512×512×8bit的超光谱图像进行压缩时,平均压缩比达到1.878,大于JPEG-LS算法和基于K-L变换的压缩比。解码后得到的重构图像的峰值信噪比为无穷大,满足超光谱图像无损压缩的要求。在TMS320DM642 EVM开发板上,使用本系统处理一幅512×512×8bit的超光谱图像,在算法优化前平均耗时为13.274s,优化后的平均耗时为7.563s,处理速度提高了43.44%,优化效果明显。更多还原

【Abstract】 In the military and civil domain, hyper-spectral image,having been applied in target classification, target identification, target tracking etc, has important research value and great application significance. With the rapid development of the remote technology and the enhancement of remote sensor resolution,the hyper-spectral image , as one of the main remote sensing data, which increased a lot, brings in serious problems about transmission and storage of hyper-spectral image data. So far, one kind of mature or standard hyper-spectral image compression technology has been not farmed. Therefore, the research on remote hyper-spectral image compression coding has great application value.Because hyper-spectral image has abundance detail and great data quantity, there always has some limits in some degree to compress the image adopting traditional compression methods,And it is much harder to achieve lossless compression about the hyper-spectral image. At present, much attention has been paid to the nearly lossless compression research. In this paper, taking the moving-scanned hyper-spectral image as the study object, Combined with the image’s unique feature caused by the special shooting method, a simple DPCM predication model is build. And choose the LeGall5/3 wavelet base to realize the integral wavelet transformation on the difference image from the model, base on integral lifting scheme The algorithm above only refer to some add and subtract and shift operations, and it can rebuild the wavelet transform coefficients perfectly. And it is easy to present on the hardware. The related program which includes image-data reading module, image-pretreatment module, spectral-prediction module, IWT module, and Embedded Zero-tree Wavelets module have been completed by using the standard C, and the very algorithm compiled for the design have passed the test by using the CCS integrated development environment and the TDS320DM642EVM development board. Finally, gathered the CCS integrated development environment and the TDS320DM642EVM development board, the algorithm has been optimized by using the compiler optimization, software assembly line technology, data-packing technology, DMA, improved disposition and storage space, etc, which have raised the utilization rate of hardware source and the usability of the system much.The experiments shows that the average compression ratio of compress a 512×512×8bit hyper-spectral image lossless is about 1.878, by using the system to compress a 512×512×8bit hyper-spectral image. The effect is better than the JPEG-LS alogorithm and the k-l transform compression algorithm to the 2-D images. The peak signal-to-noise ratio(PSNR) of reconstruction images is infinite and satisfies the image quality requirements. Using the system algorithm to deal with a 512×512×8bit hyper-spectral images, under the condition of TMS320DM642, cost about 13.274s. When optimized, the average time is about 7.563s. The processing speed is improved by 43.44%, which the optimizing effect is obvious.更多还原