The study of enhanced earth observations on a satellite image chain

The Multi-Spectral Camera (MSC) on the KOrea Multi-Propose SATellite (KOMPSAT)-2 was developed and launched as a main payload to provide a One(1) m panchromatic image and four(4) band four(4) m multi-spectral images at an altitude of 685 km covering a swath width of 15 km. These images, archived around the world, are a useful resource for space applications in agriculture, cartography, geology, forestry, regional planning, surveillance, and national security. The image quality of KOMPSAT-2 depends upon its image chain, which is comprised of an on-board system in the satellite and a processing system at the ground station. Therefore, in this study we determine the factors that have a major impact on the image quality through an investigation of the entire image chain. Consequently, two methods, involving a compression algorithm and a deconvolution technique, were determined as having a significant influence on the KOMPSAT-2 image quality. The compression algorithm of KOMPSAT-2 is rate-controlled JPEG-like algorithm that controls the mismatch between the input and output data rate. The ability to control the input/output data rate may be useful during the operation of the satellite but can also lower the overall image quality. The deconvolution technique may increase the sharpness of images, but it can also amplify the image noise level. Therefore, we propose methods of wavelet-based compression and denoising as an alternative to currently existing algorithms. Satisfactory results were obtained through experimentation with these two algorithms, and they are expected to be successfully implemented into the future KOMPSAT series to yield high-quality images for enhanced earth observation.     

Orbit determination performances using single- and double-differenced methods: SAC-C and KOMPSAT-2

In this paper, Global Positioning System-based (GPS) Orbit Determination (OD) for the KOrea-Multi-Purpose-SATellite (KOMPSAT)-2 using single- and double-differenced methods is studied. The requirement of KOMPSAT-2 orbit accuracy is to allow 1 m positioning error to generate 1-m panchromatic images. KOMPSAT-2 OD is computed using real on-board GPS data. However, the local time of the KOMPSAT-2 GPS receiver is not synchronized with the zero fractional seconds of the GPS time internally, and it continuously drifts according to the pseudorange epochs. In order to resolve this problem, an OD based on single-differenced GPS data from the KOMPSAT-2 uses the tagged time of the GPS receiver, and the accuracy of the OD result is assessed using the overlapping orbit solution between two adjacent days. The clock error of the GPS satellites in the KOMPSAT-2 single-differenced method is corrected using International GNSS Service (IGS) clock information at 5-min intervals. KOMPSAT-2 OD using both double- and single-differenced methods satisfies the requirement of 1-m accuracy in overlapping three dimensional orbit solutions. The results of the SAC-C OD compared with JPL’s POE (Precise Orbit Ephemeris) are also illustrated to demonstrate the implementation of the single- and double-differenced methods using a satellite that has independent orbit information available for validation.     

Validation of an active transponder for KOMPSAT-5 SAR image calibration

This paper describes the development and validation of a transportable active transponder designed for the image calibration of Korea Multi-Purpose Satellite-5 (KOMPSAT-5) with a synthetic aperture radar (SAR). Ground targets are essential in SAR image calibration. The environment for the deployment of ground targets for SAR image calibration should provide uniformity and minimum interference. The Amazon or deserts are regarded as desirable environments. However, such environments for SAR image calibration are difficult to find in Korea. Thus, it will be advantageous to have an active transponder whose performance will not be severely limited by the absence of such uniform environment. We have therefore developed an active transponder which has an adjustable internal delay and into which the orbit data of an arbitrary satellite can be loaded. The stored obit data with the aid of an internal global positioning system (GPS) receiver and gyroscope enables the active transponder to point to a selected satellite. In addition, a virtual deployment of the active transponder is possible due to its adjustable internal delay. Thus, the developed active transponder can be deployed at any place without environmental constraint. The performance of the developed active transponder is validated using the satellite TerraSAR-X, which is already in operation. The test results show that the active transponder is successfully compliant with the requirements for KOMPSAT-5 image calibration.     

Statistical description of low-latitude plasma blobs as observed by DMSP F15 and KOMPSAT-1

The global distribution of low-latitude plasma blobs was investigated by in-situ plasma density measurements from the Korea Multi-Purpose Satellite-1 (KOMPSAT-1) and Defense Meteorological Satellite Program (DMSP) F15. In the observations, blobs occurred in the longitude sector where the activity of the equatorial plasma bubble (EPB) was appreciable, and additional blobs were found at the lower (KOMPSAT-1) altitude as in the EPBs. However, several notable differences exist between the distributions of EPBs and blobs. First, KOMPSAT-1 found few blobs around 0°E in March and June, as did DMSP F15 from 30°W to 120°E for every season. Second, the overall occurrences in December and March at the DMSP F15 (840 km) altitude were somewhat lower than expected from those of the EBPs. Third, at the DMSP F15 altitude, the occurrence probability of plasma blobs was less controlled by yearly variations in the solar activity. These results imply that topside ionospheric conditions as well as the existence of EPBs control further development of blobs. Additionally, it was found that the blob latitudes became higher as the yearly solar activity increased. Moreover, most of the blobs were encountered in the winter hemisphere, possibly due to the low ambient density.     

Onboard orbit determination using GPS observations based on the unscented Kalman filter

Spaceborne GPS receivers are used for real-time navigation by most low Earth orbit (LEO) satellites. In general, the position and velocity accuracy of GPS navigation solutions without a dynamic filter are 25 m (1σ) and 0.5 m/s (1σ), respectively. However, GPS navigation solutions, which consist of position, velocity, and GPS receiver clock bias, have many abnormal excursions from the normal error range for space operation. These excursions lessen the accuracy of attitude control and onboard time synchronization. In this research, a new onboard orbit determination algorithm designed with the unscented Kalman filter (UKF) was developed to improve the performance. Because the UKF is able to obtain the posterior mean and covariance accurately by using the second-order Taylor series expansion through the sampled sigma points that are propagated by using the true nonlinear system, its performance can be better than that of the extended Kalman filter (EKF), which uses the linearized state transition matrix to predict the covariance. The dynamic models for orbit propagation applied perturbations due to the 40 × 40 geo-potential, the gravity of the Sun and Moon, solar radiation pressure, and atmospheric drag. The 7(8)th-order Runge–Kutta numerical integration was applied for orbit propagation. Two types of observations, navigation solutions and C/A code pseudorange, can be used at the user’s discretion. The performances of the onboard orbit determination were verified using real GPS data of the CHAMP and KOMPSAT-2 satellites. The results of the orbit determination were compared with the precision orbit ephemeris (POE) of the CHAMP and KOMPSAT-2 satellites.     

Effects of solar activity variations on the low latitude topside nighttime ionosphere

We have studied the topside nighttime ionosphere of the low latitude region using data obtained from DMSP F15, ROCSAT-1, KOMPSAT-1, and GUVI on the TIMED satellite for the period of 2000–2004, during which solar activity decreased from its maximum. As these satellites operated at different altitudes, we were able to discriminate altitude dependence of several key ionospheric parameters on the level of solar activity. For example, with intensifying solar activity, electron density was seen to increase more rapidly at higher altitudes than at lower altitudes, implying that the corresponding scale height also increased. The density increased without saturation at all observed altitudes when plotted against solar EUV flux instead of F10.7. The results of the present study, as compared with those of previous studies for lower altitudes, indicate that topside vertical scale height increases with altitude and that, when solar activity increases, topside vertical scale height increases more rapidly at higher altitudes than at lower altitudes. Temperature also increased more rapidly at higher altitudes than at lower altitudes as solar activity increased. In addition, the height of the F2 peak was seen to increase with increasing solar activity, along with the oxygen ion fraction measured above the F2 peak. These results confirm that the topside ionosphere rises and expands with increasing solar activity.     

基于图像拼接的微小零件视觉检测方法

针对毫米级微小尺寸零件的高准确度检测问题,提出了一种基于图像拼接技术的视觉检测方法,即利用亚像素微小位移结合相位相关的自动拼接方法。以两幅图像的拼接为例,这种方法首先对第一幅图像利用三次样条插值,进行亚像素微小位移;然后用处理后的第一幅图像与第二幅图像进行相位相关处理,计算两者的相关功率谱函数分布图;再根据分布图中功率谱最大值求出两幅图像的亚像素配准关系;最后根据配准关系把两幅图像拼接起来。实验表明,使用这种拼接方法,准确度可达到亚象素级,并具有抗噪声、易实现自动化等特点。     

基于分形协方差函数的自然纹理描述与分类

研究了自然纹理图像的描述与分类的方法,提出了基于分数布朗运动模型及其协方差函数的方法.分数布朗运动的协方差函数被用来估计自然纹理特征的Hurst系数和常数k.2个子图像的5个特征组成10个特征的特征集.与直接从原始纹理图像获得特征矢量不同,该方法的10个特征矢量是分别基于大于图像灰度平均值的图像和小于图像灰度平均值的图像得到的.以纹理图像的平均值为阈值,可以得到2幅子纹理图像.从每个子纹理图像提取出5个特征,它们分别是横向、纵向和45°方向的常数,横向和纵向距离为2的Hurst系数.2个子纹理的5个特征组成10个特征的特征集.从Brodatz纹理集选出的16种纹理图像被用来检验描述和分类效果,分类结果显示该方法具有很好的自然纹理的描述和分类能力.     

一种基于DCT的SAR图像自聚焦算法

研究了一种基于离散余弦变换(DCT, Discrete Cosine Transform)的SAR(Synthetic Aperture Radar)图像自聚焦算法.该方法从复图像域出发,通过在距离压缩相位历史域引入相位误差模型,改变图像的聚焦程度直至一维像的DCT序列在高序数区域的权值达到最大,从而完成误差校正.同相位梯度自聚焦算法相比,该方法无需在图像域分离出强点目标,因此特别适用于无任何明显特征的图像.由于DCT存在快速算法,使得该自聚焦算法计算量较少,更易实现.实测数据及仿真数据的成像结果证明了此方法的可行性.      

基于无监督学习的多模态可变形配准

针对医学图像配准问题，传统方法提出通过解决优化问题进行配准，但计算成本高、运行时间长。深度学习方法提出使用网络学习配准参数，从而进行配准并在单模态图像上取得高效性能。但在多模态图像配准时，不同模态图像的强度分布未知且复杂，大多已有方法严重依赖标签数据，现有方法不能完全解决此问题。提出一种基于无监督学习的深度多模态可变形图像配准框架。该框架由基于损失映射量的特征学习和基于最大后验概率的变形场学习组成，借助空间转换函数和可微分的互信息损失函数实现无监督训练。在MRI T1、MRI T2以及CT的3D多模态图像配准任务上，将所提方法与现有先进的多模态配准方法进行比较。此外，还在最新的COVID-19的CT数据上展示了所提方法的配准性能。大量结果表明:所提方法与其他方法相比，在配准精度上具有竞争优势，并且大大减少了计算时间。      

Construction of lunar DEMs based on reflectance modelling

Existing lunar DEMs obtained based on laser altimetry or photogrammetric image analysis are characterised by high large-scale accuracies while their lateral resolution is strongly limited by noise or interpolation artifacts. In contrast, image-based photometric surface reconstruction approaches reveal small-scale surface detail but become inaccurate on large spatial scales. The framework proposed in this study therefore combines photometric image information of high lateral resolution and DEM data of comparably low lateral resolution in order to obtain DEMs of high lateral resolution which are also accurate on large spatial scales. Our first approach combines an extended photoclinometry scheme and a shape from shading based method. A novel variational surface reconstruction method further increases the lateral resolution of the DEM such that it reaches that of the underlying images. We employ the Hapke IMSA and AMSA reflectance models with two different formulations of the single-particle scattering function, such that the single-scattering albedo of the surface particles and optionally the asymmetry parameter of the single-particle scattering function can be estimated pixel-wise. As our DEM construction methods require co-registered images, an illumination-independent image registration scheme is developed. An evaluation of our framework based on synthetic image data yields an average elevation accuracy of the constructed DEMs of better than 20 m as long as the correct reflectance model is assumed. When comparing our DEMs to LOLA single track data, absolute elevation accuracies around 30 m are obtained for test regions that cover an elevation range of several thousands of metres. The proposed illumination-independent image registration method yields subpixel accuracy even in the presence of 3D perspective distortions. The pixel-wise reflectance parameters estimated simultaneously with the DEM reflect compositional contrasts between different surface units. Specifically, the detected variations of the parameter of the single-particle scattering function indicate small-scale variations of the regolith particle size, possibly as a result of differences in soil maturity.     

一种图像缩放算法的SoC协同加速设计方法

针对无人机自主着陆的跑道检测、识别、跟踪等视觉算法中需要对大量图像进行缩放处理以便后续计算,但又对实时性要求比较高的情况,根据输入输出像素点的映射关系提出了一种适用于硬件加速的图像缩放算法,简化算法结构的同时利用现场可编程门阵列进行模块硬件功能的设计对算法加速,并采用软硬件协同的体系结构搭建实时图像处理系统。实验结果表明,该缩放算法处理精度高、耗时少,且用硬件逻辑实现后,可以进一步提速171倍,硬化后的系统可以通过摄像头获取图像数据,实时处理后在显示器中显示,达到30帧/s的处理速度,可以应用于实时性要求较高的图像处理算法中。      

基于DSP和FPGA技术的细胞图像采集系统设计

细胞学研究领域中需要对大量细胞的生长情况进行长期的在线跟踪、记录和分析,针对细胞图像采集和处理中的数据量大、采样频率高、运算复杂等问题,设计了一种新颖的细胞图像采集系统,讨论了DSP(Digital Signal Processor)处理系统和FPGA(Field Programmable Gate Arrays)逻辑控制系统设计中的关键技术问题,以及JPEG图像压缩算法的实现问题.系统主要由视频解码芯片、FPGA以及DSP等组成,具有功能集成、结构简单、编程灵活的特点,能够实现对大量细胞进行长期观测记录的图像采集,以及后期图像数据处理的功能.      

基于残缺图像样本的生成对抗网络图像修复方法

针对大面积图像修复缺失严重时，需要完整且高质量训练样本的问题，提出了一种将残缺或含噪图像样本作为训练集的双生成器深度卷积生成对抗网络（DGDCGAN）模型。构建两个生成器和一个鉴别器以解决单一生成器收敛慢的问题，用残缺图像样本作为训练集，通过交叉计算、搜索损失区域类似的图像信息作为训练生成模型的样本，收敛速度更快。鉴别器损失函数改进为输出的Wasserstein距离，使用自适应估计算法优化生成器损失函数和鉴别器损失函数的模型参数，最小化两两图像之间的总距离差，使用鉴别模型和修复图像总距离变化均方差最小化两个指标优化修复结果。在4个公开数据集上进行主客观实验，结果表明:所提方法能使用残缺图像样本作为训练集，有效实现大面积失真图像的修复，且收敛速度和修复效果优于现有图像修复方法。      

基于小波域清晰度评价的光场全聚焦图像融合

传统全聚焦图像融合以相机多次曝光拍摄的多聚焦图像为基础,光场相机在单次曝光后可计算空间任意深度的重聚焦图像,为后期全聚焦图像的获取提供便利。提出了一种基于小波变换的光场全聚焦图像获取算法,可有效避免传统空域图像融合算法的块效应,获得较高质量的全聚焦图像。该算法通过对微透镜阵列光场相机获得的4D光场数据进行空间变换与投影,得到用于全聚焦图像融合的重聚焦图像,对各帧重聚焦图像进行小波分解提取高、低频子图像集,提出区域均衡拉普拉斯算子、像素可见度函数分别构建融合图像的高、低频小波系数实现图像融合,其性能优于传统的区域清晰度评价函数。实验验证了所提算法的正确性和有效性,采用Lytro光场相机的原始数据计算了融合全聚焦图像,与传统图像融合算法相比,人眼视觉效果更好,客观图像指标也得到了提高。      

Identifying areas of high economic-potential copper mineralization using ASTER data in the Urumieh–Dokhtar Volcanic Belt,Iran

This study investigates the application of spectral image processing methods to ASTER data for mapping hydrothermal alteration zones associated with porphyry copper mineralization and related host rock. The study area is located in the southeastern segment of the Urumieh–Dokhtar Volcanic Belt of Iran. This area has been selected because it is a potential zone for exploration of new porphyry copper deposits. Spectral transform approaches, namely principal component analysis, band ratio and minimum noise fraction were used for mapping hydrothermally altered rocks and lithological units at regional scale. Spectral mapping methods, including spectral angle mapper, linear spectral unmixing, matched filtering and mixture tuned matched filtering were applied to differentiate hydrothermal alteration zones associated with porphyry copper mineralization such as phyllic, argillic and propylitic mineral assemblages.     

A novel satellite image encryption algorithm based on hyperchaotic systems and Josephus problem

With the rapid growth of the number of Earth observation satellite (EOS) supporting critical applications, it is required to improve the security techniques to protect the sensitive data and images during the transmission between the satellites and the ground stations. This paper introduces a new satellite image encryption algorithm based on the Linear Feedback Shift Register (LFSR) generator, SHA 512 hash function, hyperchaotic systems, and Josephus problem. LFSR generates a matrix that is used to construct the 512-bits value of the hash function. These bits are used to set the initial values and parameters of the proposed encryption algorithm. Firstly, the six dimensions (6-D) hyperchaotic system is divided into three parts, where every two equations are considered as one part. Secondly, the 1-D hyperchaotic logistic-tent system is considered as the controller to select one part. The selected part is used to generate a matrix that is XORed with the original image. Thirdly, the scrambling operation by Josephus sequences is applied to the output of the previous step by scrambling the rows and the columns according to the selected part to produce the pre-encrypted image. Finally, if the number of iterations is less than the required number which is considered as a parameter of the secret key, the previous operations will be repeated in the pre-encrypted image; otherwise, the pre-encrypted image is considered as the final cipher image. Experimental and analyses results show that the proposed algorithm has good performance in terms of high level of security, large enough key-space, tolerance to Single Event Upsets (SEU) as well as low time complexity.     

一种基于方差的自适应火星图像阈值选取算法

在基于光学成像的深空探测自主导航中,图像边缘处理直接影响视线矢量的提取,从而对自主导航的精度产生较大影响。传统的Otsu算法更侧重同区域灰度的均匀性,适用于图像中目标区域和背景区域面积相差不大的情况,因此在自主导航初始阶段,导航精度较低。根据深空探测巡航段拍摄到的火星图像的特点,基于已有火星探测任务的实拍图像,在Otsu算法的基础上,重新设计准则函数,提出了一种基于方差的火星图像阈值自适应选取算法。该算法将灰度值高于待定阈值的区域的方差表示为待定阈值的函数,以函数一阶微分的最大值对应的灰度值作为最优阈值。该方法具有计算量小、图像分割精确的优点。仿真结果表明,相较于传统的Otsu算法,通过该算法得到的图像阈值能够实现更高的视线矢量提取精度及自主导航精度。     

极稀疏投影数据的CT图像重建

从稀疏投影数据足够精确地重建断层图像，从而能够在显著降低计算机断层成像（CT）检查X-射线辐射剂量的前提下，提供充分适宜影像学临床诊断需求的重建图像。针对圆周扫描扇束投影的二维CT图像重建，提出了投影驱动的系统模型，并将CT迭代图像重建与压缩感知（CS）理论相结合，设计了一种CT迭代图像重建算法，且将算法扩展到圆周扫描锥束投影的三维CT图像重建。仿真实验结果表明:在极稀疏投影数据的条件下（[0，2π）范围内扇束/锥束扫描不超过20个投影角度），算法数值精度高，计算复杂度低，内存开销少，有很强的工程实用性。