SAR图像降噪算法工程化实现方法研究

发布时间：2018-05-02 23:23

本文选题：SAR图像 + 降噪　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：SAR图像即为合成孔径雷达图像,由于相干成像原理,SAR图像受到相干斑噪声的干扰,而相干斑噪声模型并非传统的加性噪声模型,而是难以处理的乘性噪声模型。在SAR图像的降噪处理中,一般是将成像数据传送至地面工作站,采用较为复杂且性能优良的降噪算法,利用工作站进行并行计算以得到降噪图像。然而在一些资源有限的即时应用场合,例如SAR图像的即时显示或者高倍率压缩后进行快速传输等,这些场合难以利用工作站来预先对图像进行降噪处理,而噪声的存在对该应用又存在较大影响。因此通过研究SAR图像降噪处理算法,并利用集成度高、规模小的硬件系统来实现对SAR图像的并行降噪处理,这种工程实现研究具有实际的需求和意义。本文在理解SAR图像的特点及相关滤波方法后,综合考虑降噪性能和硬件实现的复杂度,选取了基于非下采样小波变换的阈值降噪算法来对SAR图像进行降噪。该算法的原理、工程化实现方法、计算复杂度和降噪效果均在文中作了详细介绍。在此基础上设计相应的专用指令集处理器ASIP内核,并通过体系结构、指令集和流水线,对该ASIP进行完整的描述。由于处理过程中涉及对数指数运算,该部分电路也较为关键,因此对其设计的详细过程也作了描述。以ASIP构成的ASIP阵列被用来对图像进行并行降噪,ASIP阵列采用MIMD结构的组织形式,而图像被分割为图像块并交由ASIP进行处理。在明确了ASIP阵列的总体结构及数据处理方式之后,结合软硬件联合仿真进行功能化的验证,并通过XILINX ISE集成开发软件进行综合及布局布线,得到资源统计和时序分析结果。最后基于实验室的专用图像处理平台,利用FPGA实现SAR图像的降噪处理电路,得到工程化的降噪结果并进行评价。结果显示,本文的SAR图像降噪处理电路具备独特的优点,对资源限定场合的应用具有参考意义。
[Abstract]:The SAR image is a synthetic aperture radar image. Because of the principle of coherent imaging, the speckle noise model is not the traditional additive noise model, but the multiplicative noise model which is difficult to deal with. In the denoising processing of SAR image, the imaging data is usually transmitted to the ground workstation. The noise reduction algorithm with more complex and excellent performance is adopted, and the parallel computation is carried out by the workstation to get the denoising image. However, in some real-time applications with limited resources, such as real-time display of SAR images or fast transmission after high rate compression, it is difficult to use workstations to pre-process the images. However, the existence of noise has a great influence on the application. Therefore, by studying the algorithm of SAR image de-noising and using the hardware system with high integration and small scale to realize the parallel de-noising of SAR image, this engineering implementation has practical demand and significance. After understanding the characteristics of SAR images and the related filtering methods, considering the performance of noise reduction and the complexity of hardware implementation, this paper selects a threshold denoising algorithm based on non-downsampling wavelet transform to reduce the noise of SAR images. The principle, engineering implementation, computational complexity and noise reduction effect of the algorithm are introduced in detail in this paper. On this basis, the ASIP kernel of the corresponding special instruction set processor is designed, and the ASIP is described completely through the architecture, instruction set and pipeline. Because the logarithmic exponent operation is involved in the process of processing, the circuit is also very important, so the detailed design process is also described. The ASIP array composed of ASIP is used to reduce the noise of the image in parallel. The MIMD structure is used in the ASIP array, and the image is segmented into blocks and processed by ASIP. After defining the overall structure and data processing mode of ASIP array, the functional verification is carried out with the combination of software and hardware simulation, and the results of resource statistics and time series analysis are obtained by synthesizing and routing by XILINX ISE integrated development software. Finally, based on the special image processing platform of the laboratory, the noise reduction circuit of SAR image is realized by using FPGA, and the engineering denoising results are obtained and evaluated. The results show that the SAR image denoising processing circuit in this paper has unique advantages and has reference significance for the application of resource-limited occasions.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【共引文献】