摘 要

推扫式高光谱成像系统分为内部推扫式高光谱成像系统和外部推扫式高光谱成像系统，本文采用内部推扫式高光谱成像系统进行采集数据，并针对内部推扫式高光谱成像系统原理分析出实际采集图像中对焦方面可能遇到的色散问题。本文利用MATLAB处理内部推扫式高光谱成像系统所采集的图像数据，调用峰显著度函数对图像数据进行处理，通过峰显著度和，峰显著度均值，滤波后峰显著度均值等作为指标来量化评价图像清晰程度。通过视觉感官评价对比，最终确定滤波后峰显著度均值为最佳描述图像清晰度变化指标。借助该指标对高光谱成像的清晰度进行分析，确定了图像的清晰度在焦距一定的情况下随成像波段变化的规律。因此，不存在一个固定焦距适用于高光谱成像所有波段，根据使用要求确定针对感兴趣波段，确定最佳图像焦距策略与算法，即首先用MATLAB对感兴趣波段内滤波后峰显著度均值进行比较，感兴趣波段内每个波长的光谱图像的亮度曲线进行计算峰显著度数值，然后对峰显著度进行滤波均值处理，作为各波段的清晰度指标。按波段对不同焦距成像所得的滤波后均值结果进行选举，统计胜率最高的焦距值作为该感兴趣区，最佳焦距参数。

关键词：内部推扫式高光谱；自动对焦算法；峰显著度

Hyperspectral imaging system autofocus algorithm development

ABSTRACT

The push-broom hyperspectral imaging system is divided into an internal push-scan hyperspectral imaging system and an external push-sweep hyperspectral imaging system. This paper uses an internal push-sweep hyperspectral imaging system to acquire data and is used for internal push-scan hyperspectral imaging. The system principle analyzes the dispersion problems that may be encountered in the actual acquisition image. In this paper， MATLAB is used to process the image data collected by the internal push-broom hyperspectral imaging system， and the peak saliency function is used to process the image data. The peak saliency degree， the peak saliency average， and the filtered peak saliency mean are used as indicators. Quantitatively evaluate the clarity of the image. Through the visual sensory evaluation comparison， the average value of the peak significance of the filtered peak is finally determined as the best description of the image sharpness change index. With the aid of this index， the sharpness of hyperspectral imaging is analyzed， and the law of image sharpness changes with the imaging band at a certain focal length. Therefore， there is no fixed focal length suitable for all bands of hyperspectral imaging. According to the requirements of use， determine the optimal image focal length strategy and algorithm for the band of interest， that is， first compare the mean value of the filtered peak saliency in the band of interest with MATLAB. The brightness curve of the spectral image of each wavelength in the band of interest is used to calculate the peak saliency value， and then the peak saliency is filtered and averaged as the sharpness index of each band. The filtered average results obtained by imaging different focal lengths are selected according to the band， and the focal length value with the highest winning ratio is counted as the region of interest and the best focal length parameter.

Key words: internal push-scan hyperspectral;autofocus algorithm;peak saliency

目录

1 绪论 - 1 -

1.1 研究背景 - 1 -

1.2 研究内容及技术路线 - 3 -

2 内部推扫高光谱成像原理与定焦难点 - 4 -

2.1 推扫式高光谱成像系统 - 4 -

2.2 内部推扫式高光谱成像系统 - 5 -

2.3 影响定焦的因素 - 6 -

2.3.1 色散因素 - 6 -

2.3.2 场曲因素 - 7 -

3 高光谱图像数据采集 - 8 -

3.1 高光谱成像系统 - 8 -

3.2 数据采集 - 9 -

4 图像分析 - 11 -

4.1 感官评价ENVI工具 - 11 -

4.2 自动量化评价指标 - 13 -

4.2.1 量化评价工具 - 13 -

4.2.2 峰显著度和指标 - 13 -

4.2.3 峰显著度均值指标 - 21 -

4.2.4 峰显著度均值滤波指标 - 35 -

4.3 高光谱图像的定焦模型与算法 - 44 -

4.3.1 定焦模型 - 44 -

4.3.2 定焦算法 - 45 -

结论 - 49 -

致谢 - 50 -

参考文献 - 51 -

1 绪论

1.1 研究背景

高光谱成像技术是最近几十年中发展起来的基于非常多连续的窄波段的图像数据技术，它集合了光学，电子学，信息处理技术，计算机图像处理技术等领域的先进技术，集合了传统的成像技术和光谱技术的一门新兴技术^[1]。高光谱是利用很多窄的电磁波波段获取物体有关数据的技术，它可在电磁波的紫外、可见光、近红外、中红外以至热红外区域，获取许多非常窄且光谱连续的图像数据，为每个像元提供数十至数百个窄波段(通常波段宽度＜10 nm)光谱信息，能产生一条完整而连续的光谱曲线^[2]。

高光谱成像技术的定义是在多光谱成像的基础上，利用成像光谱仪，在光谱覆盖范围内的数十或数百条光谱波段对目标物体连续成像。在获得物体空间特征成像的同时，也获得了被测物体的光谱信息。高光谱成像技术具有多波段(可达上百个波段)、波段窄、光谱范围广(200-2500 nm)和图谱合一等特点。优势在于采集到的图像信息量丰富，识别度较高和数据描述模型多^[3]。