基于正交投影散度的高光谱遥感波段选择算法

由于高光谱数据的海量高维特征,对其进行降维处理成为高光谱遥感研究的一个重要问题.波段选择算法由于能够有效地保留原始数据的信息,在高光谱数据降维及后续的遥感识别与分类等方面具有明显的优越性.文章提出了一种基于正交投影散度(OPD)的波段选择方法,该方法继承了正交子空间投影(OSP)算法的特点,通过把原始数据投影到特征空间...     

A general strategy for anisotropic diffusion in MR image denoising and enhancement

Anisotropic diffusion (AD) has proven to be very effective in the denoising of magnetic resonance (MR) images. The result of AD filtering is highly dependent on several parameters, especially the conductance parameter. However, there is no automatic method to select the optimal parameter values. This paper presents a general strategy for AD filtering of MR images using an automatic parameter selection method. The basic idea is to estimate the parameters through an optimization step on a synthetic image model, which is different from traditional analytical methods. This approach can be easily applied to more sophisticated diffusion models for better denoising results. We conducted a systematic study of parameter selection for the AD filter, including the dynamic parameter decreasing rate, the parameter selection range for different noise levels and the influence of the image contrast on parameter selection. The proposed approach was validated using both simulated and real MR images. The model image generated using our approach was shown to be highly suitable for the purpose of parameter optimization. The results confirm that our method outperforms most state-of-the-art methods in both quantitative measurement and visual evaluation. By testing on real images with different noise levels, we demonstrated that our method is sufficiently general to be applied to a variety of MR images.     

一种高光谱图像波段选择的快速混合搜索算法

由于高光谱图像的高数据维和大数据量,现有的波段选择方法大多不能同时具有良好的效果和较短的计算时间。提出了一种用于高光谱图像波段选择的新方法——快速混合搜索算法。该算法将全局搜索和局部寻优有机的结合起来,能够在较短的时间内获得最佳的波段组合,用于高光谱图像的目标分类识别。快速混合搜索算法克服了传统搜索方法在高光谱图像波段选择中的缺陷,能够在提高所选波段性能的同时节省大量的运算时间。分别利用200波段和126波段的AVIRIS对其数据进行了仿真实验。实验结果表明,快速混合搜索算法在所选波段性能和计算耗时方面都获得了令人满意的效果。     

Distributed Lossless Compression Algorithm for Hyperspectral Images Based on Classification

This article addresses the problem of distributed lossless compression for hyperspectral images and proposes an effective lossless compression algorithm based on classification. First, a band selection algorithm was performed on the hyperspectral images to select those bands with considerable information. Next, the K-means algorithm was performed on those selected bands to obtain the classification map. To make full use of the spectral and spatial correlation, a multilinear regression model was introduced to construct the high-quality side information of each class within the identical block according to the classification map. Subsequently, the (n, k) linear grouping codes were employed to perform the distributed source coding for each class separately. The experimental results showed that the proposed algorithm has a competitive lossless compression performance compared with other state-of-the-art algorithms.     

基于成像光谱技术的作物杂草识别研究

杂草识别是变量喷雾和物理方法精确除草的前提。利用自主设计的地面成像光谱系统在自然环境下获取了胡萝卜幼苗以及马齿苋、牛筋草和地锦等杂草在380～760 nm波长区间的高光谱数据,通过对数据归一化消除光照条件的影响之后,运用逐步法进行波段选择,采用Fisher线性判别方法对杂草与胡萝卜幼苗进行了识别。结果表明,当把每种杂草都作为一类加以精细区分时,运用选择的8个波段建立模型对杂草和胡萝卜幼苗的识别率达85%左右;当把杂草整体作为一类与胡萝卜幼苗进行区分时,运用选择的7个波段识别率高于91%。同时为了设计低成本的杂草识别系统,通过穷举法选择最优的2和3波段组合,其中最优3波段组合对杂草胡萝卜幼苗的识别能力与逐步法选择的5个波段相当,整体识别率达89%。此外发现,红边波段对杂草有着显著的识别能力。     

高光谱图像分类的ReliefF-RFE特征选择算法构建与应用

高光谱图像具有波段连续、维数高、数据量大、相邻波段相关性强的特点,可为地物分类提供更为丰富的细节信息。但是,数据中存在大量冗余信息与噪声,在图像分类中如直接利用其所有波段特征而不进行有效分析与选择,将会导致较低的计算效率和较高的计算复杂度,分类精度亦可能随着波段维数增加而出现先增后减的“休斯（Hughes）现象”。为快速地从高达数十个甚至数百个波段的高光谱图像中提取出具有较好识别能力的特征子集,从而避免“维度灾难”,将过滤式ReliefF算法和封装式特征递归消除算法（RFE）相结合,构建了ReliefF-RFE特征选择算法,可用于高光谱图像分类的特征选择。该算法根据权重阈值,利用ReliefF算法快速剔除大量无关特征,缩小并优化特征子集的范围;利用RFE算法进一步搜索最优特征子集,将缩小范围后的特征子集中与分类器关联性小、冗余的特征进行递归筛选,进而得到分类性能最佳的特征子集。采用Indian pines数据集、Salinas-A数据集与KSC数据集等3个标准数据集作为实验数据,将ReliefF-RFE算法的应用效果与ReliefF和RFE算法进行对比。结果显示,在3个数据集中,应用ReliefF-RFE算法的高光谱图像分类平均总体精度（OA）为92.94%、F-measure为92.81%,Kappa系数为91.94%;ReliefF-RFE算法的平均特征维数是ReliefF算法的37%,而平均运算时间则是RFE算法的75%。由此表明,ReliefF-RFE算法能够在保证分类精度的同时,克服过滤式ReliefF算法无法有效减小特征之间冗余以及封装式RFE算法时间复杂度较高的缺陷,具有更为均衡的综合性能,适用于高光谱图像分类的特征选择。     

基于谱聚类与类间可分性因子的高光谱波段选择

随着遥感技术和成像光谱仪的发展,高光谱遥感图像的分辨率不断提高,其庞大的数据量在提高其遥感探测能力的同时,也给分析和处理带来了很大的困难。高光谱波段选择可以有效减少数据冗余,提高分类识别精度和处理效率。因此如何从多达数百个波段的高光谱图像中选择出具有较好分类识别能力的波段组合是亟待解决的问题。针对上述问题,采用基于图论的谱聚类算法,将原始高光谱图像中的波段作为待聚类的数据点,利用互信息描述两两波段间的相似度,生成相似度矩阵。再根据图谱划分理论,将相似度矩阵生成的非规范化图拉普拉斯矩阵进行谱分解,得到类间相似度小且类内相似度大的类簇;然后根据地物类型计算各波段的类间可分性因子,将其作为类簇内进一步选择代表性波段的参考指标,达到降维的目的;最后通过支持向量机与最小距离分类方法对波段选择后的图像分类。该方法区别于传统的无监督聚类方法,采用基于图论的谱聚类算法,并根据先验知识计算类间可分性因子来选择波段。通过与自适应波段选择算法和基于自动子空间划分的波段指数算法的对比实验,结果表明：两组实验当聚类数目达到相对最佳时,该波段选择方法支持向量机图像总分类精度达到94.08%和94.24%以上,最小距离分类图像总分类精度达到87.98%和89.09%以上,有效保留了光谱信息,提高了分类精度。     

基于K-L散度与光谱可分性距离的波段选择算法

波段选择是高光谱降维的常用手段,文中从波段选择应遵循的3个原则出发设计了一种基于信息散度与光谱可分性距离的波段选择算法。将高光谱数据中每个波段的光谱分量看作一个一维向量,使用K-L散度表示其相互之间的信息量,选出信息量大且相似性最小的波段组合;根据每个波段中不同地物光谱可分性距离的计算,得到可分性较大的波段组合;将两组波段组合取交集,即得到最优组合波段。为了验证算法的有效性,将选出的最佳3个波段进行伪彩色合成,对其进行光谱角制图分类,分类精度达到92.2%,Kappa系数为0.88.     

利用AOTF多光谱成像系统实现伪装目标的识别

针对传统分光器件存在移动部件以及不能快速实时选择波长的不足,搭建了用声光可调滤光器(AOTF)作为分光器件的多光谱成像系统。系统由光学镜头、AOTF、AOTF驱动器、CCD摄像机和图像采集系统组成。本系统能够在(500~1000)nm的光谱范围内成像。通过对AOTF的控制可以任意选择系统的光谱,从而有目的地选择具有典型目标特性的不同波段的光谱波长,形成同一目标在不同光谱波长下的不同图像。采用迷彩布、头盔以及自然花草进行多次目标特性识别试验,得到了能突出目标特性的具有典型光谱特性的图像。证实了基于AOTF的多光谱成像系统灵敏度高、体积小、无移动部件,并且能够快速实时地改变和选择光谱波段,在所成的多光谱图像中能提高目标与背景的对比度,对伪装目标有明显的探测和识别能力,能将伪装目标与背景区分开。     

Encoding Textual Information in Magnetic Resonance Imaging

The data of magnetic resonance imaging (MRI) studies include not only grayscale images, but also textual information associated with them —personal data about the patient, parameters of scanning and data processing, etc. This information is stored separately from graphic images. Therefore, the possibility for its correction and loss cannot be excluded. In this paper, the method of generation of marker information on diagnostic images is described. The marker information, as a textual analogue, is entered on the image during an MRI scan and becomes an integral part of the diagnostic material along with the images of anatomical structures. The method is realized by using the selective radiofrequency presaturation of non-scanable slices oriented perpendicularly to the scanned slices. It leads to the formation of bands of reduced signal in the areas of intersections of these slices on images. In this case, the band thicknesses are equal to the thicknesses of non-scanable slices. Different combinations of these bands (marker lines) are formed directly on images and can contain information about MRI studies. This information is determined not only by positions and angle orientations of bands, but also by their thickness, total brightness and brightness distribution in the transverse direction of these bands. The examples of introducing and positioning the marker information in conventional MRI studies are presented.     

改进粒子群优化算法的高光谱波段选择

高光谱图像具有数百个连续、狭窄的光谱带,光谱范围跨越可见光到红外光,可提供地物的精细光谱属性,对于地物材质和属性的识别分类具有重要应用价值。针对感兴趣目标选择有限的光谱波段进行传输和处理,对于提升高光谱数据处理时效性、以及设计面向特定应用的实用化光谱仪都具有重要意义。而如何结合目标特征选择最优波段成为在提升处理效率的同时保证目标识别或分类精度的必然要求。因此如何从数以百计维度的高光谱图像中选择出具有较好分类识别能力的波段子集是急需解决的问题。提出基于改进粒子群优化算法的高光谱波段选择方法,该方法区别于传统的粒子群优化算法,引入 “概率突跳特性”,并设定新解的淘汰机制,将“停滞”的新解进行淘汰,提高了算法的全局寻优性能。然后基于目标光谱特征采用了最优波段选择的优化目标函数,通过改进的粒子群优化算法求解目标函数,并将选定的波段子集反馈到支持向量机（SVM）中执行分类应用。采用两个标准的高光谱数据集（Indian Pines, Salinas）对选择出的波段子集进行分类测试,结果表明该方法相较于现有方法具有较高的分类精度,在几种方法中,传统的粒子群算法筛选出的波段效果最差;该算法筛选出的波段的分类精度最好,两个数据集的分类精度分别可以达到98.141 4%和99.084 8%。     

Enhancing the capabilities of binary phase only filter

In this paper, we demonstrate that the capabilities of a binary phase-only filter (BPOF) can be enhanced to identify targets irrespective of rotation, scale or the imaging spectral band by utilizing the concept of log-polar transform and image fusion. Till date, BPOFs have been considered to be the simplest of all filters and incapable of identifying distorted images or images of different spectral bands like the visible or infrared (IR) bands. The novelty of this work lies in the approach adopted to demonstrate that a BPOF is equally capable of distortion-invariance like any other distortion-invariant complex matched filter. This is done by suitably fusing the images of visible and IR bands and then taking the log-polar transformation of the fused image to synthesize the BPOF. A single BPOF is thus sufficient to identify (0-360)° in-plane rotated images, (50-190)% scaled images, combination of rotation and scale changes of the target, noisy image of both the visible and IR spectral bands. A further enhancement of the correlation peak intensity (CPI) is achieved by modifying this BPOF with Mexican-hat wavelet. The designed filter was implemented in the hybrid digital-optical correlation scheme. Correlation peak intensity and peak correlation energy (PCE) have been calculated as metrics of goodness of the proposed approach. Experimental results have been presented.     

Spectral band invariant wavelet-modified MACH filter

Filters synthesized with images of a specific spectral band in general fail to recognize targets in a different spectral band. In this paper, we therefore demonstrate the use of the wavelet-modified maximum average correlation height (WaveMACH) filter for automatic target recognition applications in both the visible and infrared (IR) spectral bands. As any input target appears different when imaged through two different sensors, i.e., a CCD or an IR camera, a WaveMACH filter synthesized using a CCD image shows no correlation with the image of the same target from an IR camera and vice-versa. Hence, separate filters are required to match the input targets from the two sensors. To avoid the synthesis and storage of separate filters, the images from CCD and IR camera are fused using Daubechies wavelet and then the rotation-invariant WaveMACH filter generated with the fused image. In all, 18 WaveMACH filters (each of 20° range) are required for in-plane rotation invariance in both the spectral bands for the full range of 0–360°. Computer simulation and experimental results implemented in hybrid digital–optical correlator architecture are shown for the proposed idea. The same filters have also been used to identify multiple targets in a scene. Performance measures like peak-to-sidelobe ratio (PSR), peak correlation energy (PCE) and correlation peak intensity (CPI) have been calculated as metrics of goodness.     

Infrared (IR) image synthesis method of IR real background and modeled IR target

An infrared (IR) image synthesis method is proposed for the synthesis of a real IR background and modeled IR target, used as IR signatures, as well as a band-transformation between short wave IR (SWIR), middle wave IR (MWIR), and long wave IR (LWIR) in an IR imaging system simulation. IR target images are created by the RadThermIR software, an IR signature prediction software. Individual radiances for IR signatures, corresponding to the max/min temperatures of a real IR background and modeled IR target image, are calculated with Planck’s law. First, an IR background of an arbitrary wavelength band is transformed to one of the other wavelength bands with the temperature-radiance characteristics. And then, after adjusting the gray levels of the arbitrary IR target signatures based on their radiances for the wavelength band of the transformed IR background, these IR target and background signatures can be synthesized as one image for a specific wavelength band. The experimental results show that the modeled IR target images, such as a modeled helicopter and F16, can be synthesized on the IR background images of three IR wavelength bands. And we confirmed that IR background images of the three IR wavelength bands can diversely be synthesized with the modeled IR targets as the setting temperature of the target and background, the target distance, and the field of view (FOV) arbitrarily.     

基于自适应字典选择的MCA图像修复方法

形态成分分析是一种基于稀疏模型的图像分析算法,其中心思想是根据信号组成成分的形态差异性选择两个合适的字典分别用来表示纹理部分和边缘卡通部分,具有良好的图像修复特性。传统上字典的选择需要由使用者根据图像内容人为确定。提出一种基于图像内容的自适应字典选择方法,根据最小能量在字典集合中选择最适合当前图像的字典并对图像进行修复。实验证明,该方法具有良好的图像修复性能。     

利用少量波段近红外光谱图像鉴定玉米种子纯度

为实现玉米杂交种的自动化快速分选,提出了应用少量近红外波段光对玉米种子进行成像,获取种子光谱图像并提取纹理特征来鉴定玉米杂交种纯度的方法。采集5个玉米品种的母本和杂交种在4个短波近红外波段的透射光谱图像和4个中波近红外波段的反射光谱图像,采用白板标定校正光谱图像,运用中值滤波、大津法去除噪声,从背景中分割出种子,应用灰度分布统计,灰度共生矩阵提取纹理特征,对同一粒种子拼接其在各波长处的特征数据,应用主成分分析和正交线性判别分析降维并获得子空间的最佳可分性,使用支持向量机建立透射和反射光谱图像纯度鉴定模型。透射和反射模型对5个玉米品种平均正确鉴别率均在85%以上。表明利用少量波段的近红外光谱图像鉴定玉米杂交种纯度是可行的。     

窄带快照式多光谱成像色彩还原方法

由于兼具光谱分辨和空间分辨能力, 快照式窄带多光谱成像在资源遥感、精准农业、医疗检测等领域将有广泛应用。由于该方法使用窄带成像来提高光谱分辨率及图像对比度,所获得的图像是灰度信息,失去了场景的色彩信息,不便专家对图像鉴别、评价与赏析。已有的色彩还原算法主要针对光谱波段带宽较宽或者多个波段叠加基本覆盖整个可见光谱范围等两种光谱成像仪,不适合窄带多光谱成像方法的色彩还原。该研究适合于快照式窄带多光谱成像的色彩还原方法,提出建立窄带多光谱彩色相机的概念。首先,提出两种窄带多光谱色彩还原方法：(1)基于CIE色度系统三刺激值色的,(2)基于贝尔阵列插值算法的;其次,分别应用两种算法还原快照式窄带多光谱相机所获得的植物、手臂及宫颈组织等三种代表性场景窄带多光谱灰度图像;之后,计算并比较表征两种算法所得的代表性场景彩色图像的均值、方差、熵及梯度等表征图像质量的参数数值,确定出适合快照式窄带多光谱成像的色彩还原方法;最后,对所确定的色彩还原算法进行色偏校正。实验结果表明,基于CIE三刺激值色彩还原方法比贝尔阵列插值法更适用于窄带多光谱成像颜色复原。配合使用CIE三刺激值色彩还原方法及灰度图象校正算法,从窄带光谱成像所获得的植物、手臂皮肤及宫颈组织的灰度图像所还原出的近彩色图像逼近物体真实色彩,满足人眼观察习惯。介绍了仅覆盖可见光光谱范围30%的窄带多光谱图像进行色彩还原的方法,该方法证明快照式窄带多光谱成像可以兼具光谱分辨能力,同时保持可供人主观辨识的色彩信息。所提出实现快照式窄带多光谱彩色成像的方法,有望设计不同于传统RGB相机的彩色相机实施方案。     

基于可见光的多波段偏振图像融合新算法

采用了一种新的基于小波变换的偏振图像融合算法.首先,将两个波段中的每一波段三幅偏振图像利用小波变换分解成低频和高频部分,低频的小波系数平均值作为融合后的低频系数,高频细节系数根据不同区域特征选择方法以及对应输入图像小波系数的窗口区域方差来确定融合后高频小波系数,得到一个波段一幅图像.接着,将得到的图像再进行小波分解,采用低频图像的小波系数最小值作为融合后的低频系数,高频图像根据纹理一致性测度的纹理检测确定融合规则,用来调整高频小波系数,将来自不同图像的特征与细节融合在一起,并对融合图像质量进行了对比评价.实验结果表明,融合后的偏振图像不仅反映了场景的偏振信息,而且还包含了丰富的光谱信息,目标与背景的衬比度也得到了增强,为进一步的目标检测和识别提供了便利.     

动态测量的高光谱图像压缩感知

针对高光谱图像相邻波段之间具有强光谱相关性的特点,为了提高高光谱图像压缩感知的重构效果,本文提出一种利用边缘信息设计动态测量率的压缩感知算法。首先,通过随机投影的分块压缩感知方法对每个图像块以固定测量率采样,重构出单波段图像作为其他波段的先验信息,并对其提取出图像边缘区域;然后,根据每个图像块边缘信息的丰富程度来自适应分配测量值。在固定总测量数的前提下,对不同图像块分配不同的测量次数。最后,利用分配好的测量次数对其余波段进行采集和重构。仿真结果表明,在相同总测量数情况下,本文提出的动态测量算法重构出的高光谱图像质量（PSNR）与传统固定测量压缩感知策略相比提高了1~4 dB,相比较下的重构时间也减少,在成功重构高光谱图像的基础上更增强了细节处的图像质量。