ALBERT

Description: The hierarchical image segmentation algorithm (referred to as HSEG) is a hybrid of hierarchical step-wise optimization (HSWO) and constrained spectral clustering that produces a hierarchical set of image segmentations. HSWO is an iterative approach to region grooving segmentation in which the optimal image segmentation is found at N(sub R) regions, given a segmentation at N(sub R+1) regions. HSEG's addition of constrained spectral clustering makes it a computationally intensive algorithm, for all but, the smallest of images. To counteract this, a computationally efficient recursive approximation of HSEG (called RHSEG) has been devised. Further improvements in processing speed are obtained through a parallel implementation of RHSEG. This chapter describes this parallel implementation and demonstrates its computational efficiency on a Landsat Thematic Mapper test scene.

Description: The hierarchical segmentation (HSEG) algorithm is a hybrid of hierarchical step-wise optimization and constrained spectral clustering that produces a hierarchical set of image segmentations. This segmentation hierarchy organizes image data in a manner that makes the image's information content more accessible for analysis by enabling region-based analysis. This paper discusses data analysis with HSEG and describes several measures of region characteristics that may be useful analyzing segmentation hierarchies for various applications. Segmentation hierarchy analysis for generating landwater and snow/ice masks from MODIS (Moderate Resolution Imaging Spectroradiometer) data was demonstrated and compared with the corresponding MODIS standard products. The masks based on HSEG segmentation hierarchies compare very favorably to the MODIS standard products. Further, the HSEG based landwater mask was specifically tailored to the MODIS data and the HSEG snow/ice mask did not require the setting of a critical threshold as required in the production of the corresponding MODIS standard product.

Description: Imaging spectroscopy, also known as hyperspectral imaging, has been transformed in less than 30 years from being a sparse research tool into a commodity product available to a broad user community. Currently, there is a need for standardized data processing techniques able to take into account the special properties of hyperspectral data. In this paper, we provide a seminal view on recent advances in techniques for hyperspectral image processing. Our main focus is on the design of techniques able to deal with the highdimensional nature of the data, and to integrate the spatial and spectral information. Performance of the discussed techniques is evaluated in different analysis scenarios. To satisfy time-critical constraints in specific applications, we also develop efficient parallel implementations of some of the discussed algorithms. Combined, these parts provide an excellent snapshot of the state-of-the-art in those areas, and offer a thoughtful perspective on future potentials and emerging challenges in the design of robust hyperspectral imaging algorithms

Description: An algorithm is developed to automatically screen the outliers from massive training samples for Global Land Survey - Imperviousness Mapping Project (GLS-IMP). GLS-IMP is to produce a global 30 m spatial resolution impervious cover data set for years 2000 and 2010 based on the Landsat Global Land Survey (GLS) data set. This unprecedented high resolution impervious cover data set is not only significant to the urbanization studies but also desired by the global carbon, hydrology, and energy balance researches. A supervised classification method, regression tree, is applied in this project. A set of accurate training samples is the key to the supervised classifications. Here we developed the global scale training samples from 1 m or so resolution fine resolution satellite data (Quickbird and Worldview2), and then aggregate the fine resolution impervious cover map to 30 m resolution. In order to improve the classification accuracy, the training samples should be screened before used to train the regression tree. It is impossible to manually screen 30 m resolution training samples collected globally. For example, in Europe only, there are 174 training sites. The size of the sites ranges from 4.5 km by 4.5 km to 8.1 km by 3.6 km. The amount training samples are over six millions. Therefore, we develop this automated statistic based algorithm to screen the training samples in two levels: site and scene level. At the site level, all the training samples are divided to 10 groups according to the percentage of the impervious surface within a sample pixel. The samples following in each 10% forms one group. For each group, both univariate and multivariate outliers are detected and removed. Then the screen process escalates to the scene level. A similar screen process but with a looser threshold is applied on the scene level considering the possible variance due to the site difference. We do not perform the screen process across the scenes because the scenes might vary due to the phenology, solar-view geometry, and atmospheric condition etc. factors but not actual landcover difference. Finally, we will compare the classification results from screened and unscreened training samples to assess the improvement achieved by cleaning up the training samples. Keywords:

Description: This document is the proceedings from the 'Science Information Management and Data Compression Workshop,' which was held on September 26-27, 1994, at the NASA Goddard Space Flight Center, Greenbelt, Maryland. The Workshop explored promising computational approaches for handling the collection, ingestion, archival and retrieval of large quantities of data in future Earth and space science missions. It consisted of eleven presentations covering a range of information management and data compression approaches that are being or have been integrated into actual or prototypical Earth or space science data information systems, or that hold promise for such an application. The workshop was organized by James C. Tilton and Robert F. Cromp of the NASA Goddard Space Flight Center.

Description: This document is the proceedings from the 'Science Information Management and Data Compression Workshop,' which was held on October 26-27, 1995, at the NASA Goddard Space Flight Center, Greenbelt, Maryland. The Workshop explored promising computational approaches for handling the collection, ingestion, archival, and retrieval of large quantities of data in future Earth and space science missions. It consisted of fourteen presentations covering a range of information management and data compression approaches that are being or have been integrated into actual or prototypical Earth or space science data information systems, or that hold promise for such an application. The Workshop was organized by James C. Tilton and Robert F. Cromp of the NASA Goddard Space Flight Center.

Description: This document is the proceedings from the fourth annual 'Space and Earth Science Data Compression Workshop,' which was held on April 2, 1994, at the University of Utah in Salt Lake City, Utah. This workshop was held in cooperation with the 1994 Data Compression Conference, which was held at Snowbird, Utah, March 29-31 1994. The Workshop explored opportunities for data compression to enhance the collection and analysis of space and Earth science data. It consisted of 13 papers presented in 4 sessions. The papers focus on data compression research that is integrated into, or has the potential to be integrated into, a particular space and/or Earth science data information system. Presenters were encouraged to take into account the scientist's data requirements, and the constraints imposed by the data collection, transmission, distribution, and archival system.

Description: A new spectral-spatial method for classification of hyperspectral images is proposed. The HSegClas method is based on the integration of probabilistic classification and shape analysis within the hierarchical step-wise optimization algorithm. First, probabilistic support vector machines classification is applied. Then, at each iteration two neighboring regions with the smallest Dissimilarity Criterion (DC) are merged, and classification probabilities are recomputed. The important contribution of this work consists in estimating a DC between regions as a function of statistical, classification and geometrical (area and rectangularity) features. Experimental results are presented on a 102-band ROSIS image of the Center of Pavia, Italy. The developed approach yields more accurate classification results when compared to previously proposed methods.

Description: The Visible Infrared Imager Radiometer Suite (VIIRS) instrument onboard the Suomi National Polarorbiting Partnership (SNPP) satellite was launched on 28 October 2011. The VIIRS has 5 imagery spectral bands (I-bands), 16 moderate resolution spectral bands (M-bands) and a panchromatic day/night band (DNB). Performance of the VIIRS spatial response and band-to-band co-registration (BBR) was measured through intensive pre-launch tests. These measurements were made in the non-aggregated zones near the start (or end) of scan for the I-bands and M-bands and for a limited number of aggregation modes for the DNB in order to test requirement compliance. This paper presents results based on a recently re-processed pre-launch test data. Sensor (detector) spatial impulse responses in the scan direction are parameterized in terms of ground dynamic field of view (GDFOV), horizontal spatial resolution (HSR), modulation transfer function (MTF), ensquared energy (EE) and integrated out-of-pixel (IOOP) spatial response. Results are presented for the non-aggregation, 2-sample and 3-sample aggregation zones for the I-bands and M-bands, and for a limited number of aggregation modes for the DNB. On-orbit GDFOVs measured for the 5 I-bands in the scan direction using a straight bridge are also presented. Band-to-band co-registration (BBR) is quantified using the prelaunch measured band-to-band offsets. These offsets may be expressed as fractions of horizontal sampling intervals (HSIs), detector spatial response parameters GDFOV or HSR. BBR bases on HSIs in the non-aggregation, 2-sample and 3-sample aggregation zones are presented. BBR matrices based on scan direction GDFOV and HSR are compared to the BBR matrix based on HSI in the non-aggregation zone. We demonstrate that BBR based on GDFOV is a better representation of footprint overlap and so this definition should be used in BBR requirement specifications. We propose that HSR not be used as the primary image quality indicator, since we show that it is neither an adequate representation of the size of sensor spatial response nor an adequate measure of imaging quality.

Description: We are engaged in a project to produce a 30m impervious cover data set of the entire Earth for the years 2000 and 2010 based on the Landsat Global Land Survey (GLS) data set. The GLS data from Landsat provide an unprecedented opportunity to map global urbanization at this resolution for the first time, with unprecedented detail and accuracy. Moreover, the spatial resolution of Landsat is absolutely essential to accurately resolve urban targets such as buildings, roads and parking lots. Finally, with GLS data available for the 1975, 1990, 2000, and 2005 time periods, and soon for the 2010 period, the land cover/use changes due to urbanization can now be quantified at this spatial scale as well. Our approach works across spatial scales using very high spatial resolution commercial satellite data to both produce and evaluate continental scale products at the 30m spatial resolution of Landsat data. We are developing continental scale training data at 1m or so resolution and aggregating these to 30m for training a regression tree algorithm. Because the quality of the input training data are critical, we have developed an interactive software tool, called HSegLearn, to facilitate the photo-interpretation of high resolution imagery data, such as Quickbird or Ikonos data, into an impervious versus non-impervious map. Previous work has shown that photo-interpretation of high resolution data at 1 meter resolution will generate an accurate 30m resolution ground reference when coarsened to that resolution. Since this process can be very time consuming when using standard clustering classification algorithms, we are looking at image segmentation as a potential avenue to not only improve the training process but also provide a semi-automated approach for generating the ground reference data. HSegLearn takes as its input a hierarchical set of image segmentations produced by the HSeg image segmentation program [1, 2]. HSegLearn lets an analyst specify pixel locations as being either positive or negative examples, and displays a classification of the study area based on these examples. For our study, the positive examples are examples of impervious surfaces and negative examples are examples of non-impervious surfaces. HSegLearn searches the hierarchical segmentation from HSeg for the coarsest level of segmentation at which selected positive example locations do not conflict with negative example locations and labels the image accordingly. The negative example regions are always defined at the finest level of segmentation detail. The resulting classification map can be then further edited at a region object level using the previously developed HSegViewer tool [3]. After providing an overview of the HSeg image segmentation program, we provide a detailed description of the HSegLearn software tool. We then give examples of using HSegLearn to generate ground reference data and conclude with comments on the effectiveness of the HSegLearn tool.