基于Zynq的多光谱遥感图像的预处理

发布时间：2018-10-09 12:17

【摘要】：静止气象卫星属于一种专门的遥感卫星,负责全天候的气象观测工作。根据它拍摄的卫星云图,可以对灾害性天气进行监测与预警,保障人民群众的生命及财产安全。由于多光谱遥感图像数据量巨大,因此对多波段的红外图像进行分类前的压缩、融合预处理,对于提高云图的分辨率,缓解传输压力,增强夜间探测能力,具有现实意义。通过对多光谱遥感图像的特性分析,发现各波段的红外云图具有一定的相关性,且分辨率低、边缘模糊。从数据的压缩、融合角度考虑,本文选择K-L变换及小波变换两种方法进行预处理研究。通过Matlab编程实现两种变换算法,综合图像的视觉效果和定量指标,比较得出K-L变换能够有效地提取相关性较低的云朵信息,且与同时段可见光云图最接近;小波变换则更多地保留源图像细节信息,适用于预处理后的云图结构分析。因此,K-L变换更适合于红外云图的预处理。本文是在Xilinx公司的Zynq平台上,采用软硬件协同设计的方法实现多通道图像数据的K-L变换。在K-L IP核的设计过程中,采用C语言编程实现K-L变换算法,经VivadoHLS工具综合后,优化生成相应的IP核。经过C/RTL协同仿真验证可知,由C语言转化的RTL级代码准确无误,最后完成IP核的封装工作。完成K-L IP核的设计后,在Zedboard开发板上进行测试。基于Zynq的多光谱遥感图像预处理系统包含图像的采集、数据的缓存、算法的处理以及预处理后数据的传输四部分。经过实验验证,采用K-L变换对三幅分辨率为640×480的图像进行一次预处理操作,PC纯软件大约需要5000毫秒,而Zynq软硬件协同实现仅需要91毫秒,速度大幅提升。基于Zynq的多光谱遥感图像的预处理技术,具有实时、高效、便利的优点,在航空航天领域的研究和应用中具有广阔的前景。
[Abstract]:Geostationary meteorological satellite is a special remote sensing satellite, which is responsible for weather observation all-weather. According to the satellite cloud image it can monitor and warn the disastrous weather and ensure the safety of people's life and property. Because of the huge amount of multispectral remote sensing images, it is of practical significance to compress and fuse the multi-band infrared images before classification, to improve the resolution of cloud images, to alleviate the transmission pressure and to enhance the nocturnal detection ability. By analyzing the characteristics of multispectral remote sensing images, it is found that the infrared cloud images of each band have certain correlation, and the resolution is low, and the edges are blurred. From the point of view of data compression and fusion, two preprocessing methods, K-L transform and wavelet transform, are selected in this paper. Two transform algorithms are realized by Matlab programming, which synthesizes the visual effect and quantitative index of the image. It is concluded that K-L transform can effectively extract the cloud information with low correlation, and is closest to the visible cloud image at the same time. Wavelet transform retains more details of source image, and it is suitable for cloud image structure analysis after preprocessing. Therefore, K-L transform is more suitable for infrared cloud image preprocessing. In this paper, the K-L transform of multi-channel image data is realized on the Zynq platform of Xilinx Company by using the method of hardware and software co-design. In the design process of K-L IP kernel, the K-L transform algorithm is realized by C language programming. After the synthesis of VivadoHLS tools, the corresponding IP kernel is optimized. The C/RTL cosimulation results show that the RTL code transformed by C language is accurate and the encapsulation of IP core is finished. After the design of K-L IP core is completed, the test is carried out on the Zedboard development board. The multispectral remote sensing image preprocessing system based on Zynq includes four parts: image acquisition, data cache, algorithm processing and data transmission after preprocessing. The experimental results show that it takes about 5000 milliseconds to preprocess three images with 640 脳 480 resolution by K-L transform, while it takes 91 milliseconds for Zynq hardware / software co-implementation. Multi-spectral remote sensing image preprocessing technology based on Zynq has the advantages of real-time, high efficiency and convenience, and has a broad prospect in the research and application of aerospace.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【相似文献】