Dr. Kerekes has worked throughout his career on advancing the state of the
art and practice of remote sensing technology through theoretical investigations,
data analyses, and modeling of remote sensing systems and their performance.
His interest has been in viewing the end-to-end remote sensing process as
a system with application performance as the system metric. Developing models
with this perspective has improved understanding of parameter sensitivities
and requirements for system design and operation. His work has emphasized
the use of statistical parametric models in propagating the information
bearing characteristics of the scene through the effects of the remote sensing
process. He has applied this approach to the study of multispectral remote
sensing systems designed for surface land cover classification, the vertical
profiling of atmospheric temperature and water vapor, and for unresolved
(sub-pixel) object detection and identification.
Current Research Projects
Dr. Kerekes is involved with a number of research projects with students, faculty, staff, and outside collaborators. The following summarizes efforts for which he is the Principal Investigator.
Multi-band Uncooled Radiometer Imager (MURI) (2017-2020). This project is supporting DRS Technologies in the development of an airborne multispectral thermal infrared imaging instrument. Our role includes science support, performance modeling, and field validation.. This work was sponsored by the National Aeronautics and Space Administration.
Spaceborne Imaging System Modeling and Simulation (2016-2019). This project is modeling and predicting performance of advanced imaging systems for use on small satellites for earth observation applications. This work was sponsored by the Lawrence Livermore National Laboratory.
Global Surveillance Augmentation Using Commercial Satellite Imaging Systems (2016-2018). This project involes the use of advanced simulation techniques to create labelled imagery for training deep learning networks used in the automated analysis of commercial satellite imagery. This work is in collaboration with a small business.
System Engineering and Architecture Studies for Future Landsat Satellites (2015-2019). This project is exploring performance sensitivity of Landsat-type instruments in the context of various system concepts and design configurations. This work was sponsored by the National Aeronautics and Space Administration.
Computer Vision Techniques Applied to High Resolution Oblique Aerial Imagery (2014-2017). This project involves the application of modern computer vision techniques to very high resolution oblique aerial imagery. Applications studied include debris mapping, building detection, and roof condition assessment. This work is in collaboration with a small business.
Past Research Projects
Interdisciplinary Advancement of the Theoretical Basis for Lidar Sensing of the Earth (2011-2015). Together with researchers at SUNY Buffalo this project is investigating the theory and phenomenology of airborne and satellite lidar system measuring the topography and condition of complex earth surfaces. Examples of complex surfaces include ice sheets with irregular crevices and forests in mountainous terrain. This work was sponsored by the National Aeronautics and Space Administration.
Modeling and Processing of Longwave Infrared Spectra of Contaminated surfaces (2011-2015). This project involves the extension of a physics-based ray-tracing approach to modeling the interactions between liquids and natural/man-made surfaces in the thermal infrared. Additional effort is focused on processing of longwave spectra collected by an imaging spectrometer. This work was in collaboration with a small business.
Multimodal Image Generation and Exploitation Study (2010-2014). This project is developing a complex, large area, urban scene simulation that will be rendered as visible/infrared hyperspectral and lidar imgery. This work was supported by the Raytheon Corporation.
Three-dimensional Geometry Extraction from Oblique Image Data (2011-2014). The objective of this research is to (semi) automatically extract 3D geometry and information about surface materials from oblique airborne visible imagery. It was joint with and sponsored by Pictometry International Corporation.
Device, Algorithm and Integrated Modeling Research for Performance-driven Multi-modal Optical Sensors (2008-2012). This effort is developing advanced multi-modal sensor concepts, models and algorithms for the adaptive sensing of targets in a complex environment. It is a joint project with researchers at Numerica Corporation. This project was supported by the Air Force Office of Scientific Research.
Phenomenology Study of Feature Aided Tracking of Dismounts in a Cluttered Urban Environment (2010-2012). The objective of this research is to understand the phenomenology of hyperspectral imaging of dismounts, determining the contrast and ability to distinquish and track individuals in an urban area. This work was supported by the Air Force Research Laboratory's Sensors Directorate.
Vehicle Tracking with Hyperspectral Imagery. In this project we are working with the Air Force Research Laboratory Sensor's Directorate to investigate the feasibility of using an airborne hyperspectral measurement of a given vehicle to find the identical vehicle in a hyperspectral image taken at a later time. To investigate this topic, we are using airborne imagery collected by RIT's Modular Imaging Spectrometer Instrument (MISI). These empirical measurements are being extended through modeling analyses using the image simulation software DIRSIG and the spectral imaging system analysis model FASSP. A general overview of this project can be found in an article in the November 2005 issue of Military Geospatial Technology.
Lidar Image Data
June 24 1609 MISI Image Data File
June 24 1609 MISI Image ENVI Header
June 24 1624 MISI Image Data File
June 24 1624 MISI Image ENVI Header
August 24 WASP RGB Image GeoTIFF Data File
August 24 WASP RGB Image ENVI Header
Urban Hyperspectral Image Data
Urban Hyperspectral Image Header
Urban Truth Image Data
Urban Truth Image Header
Urban Truth Material Index
Network-centric Urban Vigilance: Phase II. This project is investigating the engineering design and processing algorithm issues for the use of multiple hyperspectral imagers to track objects of interest in an urban environment. It is a collaboration with Gitam Technologies, Inc.
Hyperspectral Detection of Chemical and Biological Agents Using Biosensors: Phase II. The objective of this project is to investigate the use of hyperspectral imaging cameras to detect phenomenological changes in plants that have been genetically engeineered to respond to particular hazardous chemicals. It is a collaboration with researchers at Gitam Technologies, Inc., and Colorado State University.
Modeling of Enivornment Effects in Optical Scene Simulations. The DIRSIG image simulation code is a sophisticated, first principles, model for creating realistic and radiometrically correct panchromatic, thermal, multispectral, and hyperspectral images. Currently, however, it lacks the capability of adding environmental effects onto the sufaces in the simulation. Examples of environmental effects include soil and dust that has blown on to man-made surfaces, as well as water pooling on surfaces after a rain. These effects are being investigated under this project to add an environmental effects capability to DIRSIG. Measurements of the spectral reflectance and emissivity for soils and water are being made and their effects on scene spectral radiance incorporated into DIRSIG through theoretical and empirical models.
Semi-automated Generation of 3D Scenes from Multiple Remote Sensing Data Sources. The definition and construction of scenes within the DIRSIG tool is generally a manual process and can be quite labor intensive. Some large scenes have taken over a year to construct. With the increasing availability of remotely sensed data from multiple modalities, it is now possible to consider a system to automate the construction of a DIRSIG scene. Three dimensional geometries can be extracted from lidar data, material properties can be extracted from hyperspectral imagery, and small spatial details can be found from high resolution imagery. This project is exploring the issues with processing these data to extract the necessary information to provide an input scene to DIRSIG in the propoer format. It is being sponsored by the National Geospatial-Intelligence Agency (NGA) under an NGA University Research Initiative (NURI) grant.
Quantification of the Accuracy of Remote Material Identification from Airborne Hyperspectral Imagery. Many studies have examined the ability of hyperspectral imaging systems to detect full and sub-pixel objects in a scene. The material identification capability of hyperspectral imagery has been highly touted, but has received much less investigation. Partly this is because the target detection problem is much more clear; an object is detected or not. Material "identification" is much more subtle since it can be interpreted at many levels ranging from a gross classification (grass, road, trees, etc.) to uniquely associating an observed pixel in an image to a particular object. This project is exploring these questions through analysis of airborne imagery of RIT's campus collected by RIT's Modular Imaging Spectromater Instrument (MISI). These empirical analyses are being extended through mo deling analyses using the image simulation software
the spectral imaging system analysis model FASSP.
Investigating Models for Predicting the Spectral Quality of a Hyperspectral Image for Target Detection Applications. A general predicitve metric for the quality and utility of hyperspectral imagery remains a research goal for the spectral imaging community. Our work has explored some of the issues associated with such a metric in the context of target detection and terrain classification. This project, in partnership with ITT Industries Space Systems Division, is extending some of that work by comparing model-based predictions of quality to the results of spectral analysis by human analysts. An area of an urban scene with targets of interest embedded is being simulated with
DIRSIG and provided to the analysts with a challenging of finding the targets of interest. The results of these analyses will be compared to those predicted by spectral utility equations feed by model results obtained using
the spectral imaging system analysis model FASSP.
Fusion of Hyperspectral and Synthetic Aperture Radar Imagery. Hyperspectral and SAR imagery form a complementary pair of technologies for the remote sensing of land surfaces. Together, they span the ends of the electromagnetic spectrum that can be used for high resolution imaging of the land. Materials that respond similarly in one modality, can respond differently in the other. Thus, there is motivation for combined analysis for a given scene. This project is using data collected by NASA JPL's AIRSAR sensor together with hyperspectral imagery collected by the HyMap sensor over a common area in and adjacent to Yellowstone National Park in the western United States. The project is investigating the issues associated with co-registration of these two modalities in a mountainous terrain, along with various ways to combine the data and extract information about the land surface.
Investigating the Characterization of Skin Anomalies with the use of Hyperspectral Imaging. The application of hyperspectral imaging to the remote characterization of the earth's surface has been well-investigated and numerous algorithms have been developed to process the data. An interesting question arises from this work. Can these techniques be applied in a clinical setting to aid in the diagnosis of human skin anomalies using office-scale versions of hyperspectral imagers? This project, is investigating the application of spectral image processing techniques to theparticular problem of mapping the blood oxygen saturated fraction at high spatial resolution. The distribution of oxygenated hemoglobin vs. de-oxygenated hemoglobin can aid in the identification of skin areas undergoing anomalous changes.
Automated Processing of Color Images for Skin Condition Assessment. This project is developing algorithms to process color images of human skin for automated categorization of the dermatological skin condition.
Return to Dr. Kerekes' Home Page
Last updated: 5 August 2017