July 7, 2016 at 10:00am - PhD Thesis Defense - Leidy P. Dorado-Munoz - Spectral Target Detection using Schroedinger Eigenmaps

July 7, 2016 at 10:00am
Leidy P. Dorado-Munoz
Spectral Target Detection using Schroedinger Eigenmaps
PhD Thesis Defense



Hyperspectral imagery (HSI) as the output of an optical remote sensing process reflects the information about properties of objects and materials on the earth's surface.  Applications include environmental monitoring, meteorology, mapping, surveillance, etc. Many of these tasks include the finding and detection of specific objects, usually few or small that are surrounded by other materials or objects that clutter the scene and hide the relevant information. This target detection process has been boosted lately by the use of HSI. Non-linear transformation methods, mainly based on manifold learning algorithms, have shown a potential use in HSI transformation, dimensionality reduction and classification. The reduction processes includes the transformation of the image data to a new space where the relevant and hidden information is more easily revealed. These non-linear methods perform dimensionality reduction while preserving the local structure of the data, causing a minimal loss of relevant information.

One of these non-linear methods is the Schroedinger Eigenmaps (SE) algorithm, which is based on the well-known Laplacian Eigenmaps (LE), and it has been introduced as a technique for semi-supervised classification. Both algorithms, LE and SE, include the creation of an adjacency graph as a means to represent the spectral connectivity of the data set, and the eigendecomposition of a significant operator that for SE is the Schroedinger operator. The Schroedinger operator includes by definition a potential term that gives the option to encode prior information about the materials present in a scene, and steers the transformation in some convenient directions in order to cluster similar pixels together. The use of the SE algorithm is proposed in this thesis as a basis for a target detection methodology that does not require assuming any statistical or geometric models, but that boosts the separability between the class of interest and the other classes present in the image. This is performed by taking advantage of the privileged location that target pixels would have in the Schroedinger space. The proposed methodology does not just include the transformation of the data to a lower dimensional space but also includes the definition of a detector that capitalizes on the theory behind SE.  In addition, a knowledge propagation scheme is used to combine spectral and spatial information as a means to propagate the “potential constraints” to nearby points connected via the adjacency graphs.  The hope is that the propagation scheme helps to reinforce weak connections and improve the separability between most of the target pixels and the background.

May 17, 2016 at 10:00am - PhD Dissertation Defense - Fan Wang - Understanding high resolution aerial imagery using computer vision techniques

Bldg. (76) - Room 1275 (Fishbowl)
May 17, 2016 at 10:00am
Fan Wang
Understanding high resolution aerial imagery using computer vision techniques
PhD Dissertation Defense



Computer vision could make important contribution to the analysis of remote sensing or aerial imagery. However, the resolution of early satellite imagery is not sufficient to provide useful spatial features. The situation is changing with the advent of very-high-spatial-resolution (VHR) imaging sensors. The change makes it possible to use computer vision techniques to analysis of man-made structures. Meanwhile, the development of multi-view imaging techniques will allow the generation of accurate point clouds as ancillary knowledge.


This dissertation aims at developing computer vision algorithms for the high resolution aerial imagery analysis in the context of application problems including debris detection, building detection and roof condition assessment. High resolution aerial imagery and point clouds are provided by Pictometry International for this study. 


Debris detection is needed for effective debris removal and allocation of limited resources. Significant advances in aerial image acquisition have greatly enabled the possibilities for rapid and automated detection of debris. In this dissertation, a robust debris detection algorithm is proposed. Large scale aerial image is partitioned into homogeneous regions by interactive segmentation. Debris areas are identified based on extracted texture feature.


Robust building detection is an important part of high resolution aerial imagery understanding. This dissertation develops a 3D scene classification algorithm for building detection using point clouds derived from multi-view imagery. Point clouds are divided into point clusters using Euclidean Clustering. Individual point clusters are identified based on extracted spectral and 3D structure features.


The inspection of roof condition is an important step of damage claim processing in the insurance industry. Automated roof condition assessment from remotely sensed images is proposed in this dissertation. Texture classification and bag-of-words model are performed to assess the roof condition using features derived from the whole rooftop. However, considering the complexity of residential rooftop, a more sophisticated method is proposed to divide the task into two stages: 1) roof segmentation, followed by 2) classification of segmented roof regions. 

Contributions of this study include the development of algorithms for debris detection using 2D images and building detection using 3D point clouds. For roof condition assessment, the solutions to this problem are explored in two directions: features derived from the whole rooftop and features extracted from each roof segments. Through our research, roof segmentation followed by segments classification is found to be a more promising method and the workflow processing is developed and tested. Since the methodology to solve these problems is focused on the design of hand-crafted features, unsupervised feature extraction techniques using deep learning should be explored in future work.




Carlson DIRS Lab 76-3215
May 16, 2016 at 10:00am
Ph. D Thesis Defense

Vehicle tracking from an aerial platform poses a number of unique challenges including the small number of pixels representing a vehicle, large camera motion, and parallax error. For these reasons, it is a more challenging task than  traditional object tracking and is generally tackled through a number of different sensor modalities. Recently, the Wide Area Motion Imagery (WAMI) sensor platform has received considerable attention as it can provide higher resolution single band imagery in addition to its large area coverage. Despite these advantages, there is still not enough feature information and most WAMI systems struggle to persistently track vehicles. Additional modalities, such as spectral data, can be cruical in identifying objects even in low resolution scenes and advances in sensor technology is starting to make hyperspectral data acquisition at video frame rates possible. For this reason, a multi-modal optical sensor concept is considered in this thesis to improve tracking in adverse scenes.


The sensor considered is based on the Rochester Institute of Technology Multi- object Spectrometer, which is capable of collecting limited hyperspectral data at desired locations in addition to full-frame single band imagery. By acquiring hyperspectral data quickly, tracking can be achieved at reasonable frame rates which is crucial for tracking. More spectral samples can lead to a huge volume of data, so the relatively high cost of hyperspectral data acquisition and transmission need to be taken into account to design a realistic tracking system. By collecting and analyzing the extended (spectral) data only for the pixels of interest, we can address or avoid the unique challenges posed by aerial tracking. To accomplish this, we integrate limited hyperspectral data to improve measurement-to-track association. Also, a hyperspectral data based target detection method is presented to avoid the parallax effect and reduce the clutter density. Finally, the proposed system is evaluated on realistic, synthetic  scenarios generated by the Digital Image and Remote Sensing Image Generation software.

April 28, 2016 at 9:00am - Ph. D Thesis Defense - SIYU ZHU - Text Detection in Natural Scenes and Technical Diagrams with Convolutional Feature Learning and Cascaded Classification

Carlson Bldg. (76) - Room 1275 (Fishbowl)
April 28, 2016 at 9:00am
Text Detection in Natural Scenes and Technical Diagrams with Convolutional Feature Learning and Cascaded Classification
Ph. D Thesis Defense



An enormous amount of digital images are being generated and stored every day. Understanding text in these images is an important challenge with large impacts for academic, industrial and domestic applications. Recent studies address the difficulty of separating text targets from noise and background, all of which vary greatly in natural scenes. To tackle this problem, we develop a text detection system to analyze and utilize visual information in a data driven, automatic and intelligent way.


The proposed method incorporates features learned from data, including patch-based coarse-to-fine detection (Text-Conv), connected component extraction using region growing, and graph-based word segmentation (Word-Graph). Text-Conv is a sliding window-based detector, with convolution masks learned using the Convolutional k-means algorithm (Coates et. al, 2011). Unlike convolutional neural networks (CNNs), a single vector/layer of convolution mask responses are used to classify patches. An initial coarse detection considers both local and neighboring patch responses, followed by refinement using varying aspect ratios and rotations for a smaller local detection window. Different levels of visual detail from ground truth are utilized in each step, first using constraints on bounding box intersections, and then a combination of bounding box and pixel intersections. Combining masks from different Convolutional k-means initializations, e.g., seeded using random vectors and then support vectors improves performance. The Word-Graph algorithm uses contextual information to improve word segmentation and prune false character detections based on visual features and spatial context. Our system obtains pixel, character, and word detection f-measures of 93.14%, 90.26%, and 86.77% respectively for the ICDAR 2015 Robust Reading Focused Scene Text dataset, out-performing state-of-the-art systems, and producing highly accurate text detection masks at the pixel level.


To investigate the utility of our feature learning approach for other image types, we perform tests on 8-bit greyscale USPTO patent drawing diagram images. An ensemble of Ada-Boost classifiers with different convolutional features (MetaBoost) is used to classify patches as text or background. The Tesseract OCR system is used to recognize characters in detected labels and enhance performance. With appropriate pre-processing and post-processing, f-measures of 82% for part label location, and 73% for valid part label locations and strings are obtained, which are the best obtained to-date for the USPTO patent diagram data set used in our experiments.


To sum up, an intelligent refinement of convolutional k-means-based feature learning and novel automatic classification methods are proposed for text detection, which obtain state-of-the-art results without the need for strong prior knowledge. Different ground truth representations along with features including edges, color, shape and spatial relationships are used coherently to improve accuracy. Different variations of feature learning are explored, e.g. support vector-seeded clustering and MetaBoost, with results suggesting that increased diversity in learned features benefit convolution-based text detectors.





April 6, 2016 at 1:00am
PhD Dissertation Defense



Phase-only optical elements can provide a number of important functions for high-contrast imaging. This thesis presents analytical and numerical optical design methods for accomplishing specific tasks, the most significant of which is the precise suppression of light from a distant point source. Instruments designed for this purpose are known as coronagraphs. Here, advanced coronagraph designs are presented that offer improved theoretical performance in comparison to the current state-of-the-art. Applications of these systems include the direct imaging and characterization of exoplanets and circumstellar disks with high sensitivity. Several new coronagraph designs are introduced and, in some cases, experimental support is provided.


In addition, two novel high-contrast imaging applications are discussed: the measurement of sub-resolution information using coronagraphic optics and the protection of sensors from laser damage. The former is based on experimental measurements of the sensitivity of a coronagraph to source displacement. The latter discussion presents the current state of ongoing theoretical work. Beyond the mentioned applications, the main outcome of this thesis is a generalized theory for the design of optical systems with one of more phase masks that provide precise control of radiation over a large dynamic range, which is relevant in various high-contrast imaging scenarios. The optimal phase masks depend on the necessary tasks, the maximum number of optics, and application specific performance measures. The challenges and future prospects of this work are discussed in detail.


Carlson DIRS Lab 76-3215
March 14, 2016 at 2:30am
PhD Dissertation Defense



The holy grail of brain imaging is the identification of a biomarker, which can identify an abnormality that can be used both, to diagnose disease and track the effectiveness of treatment and disease progression. Typically approaches that search for biomarkers start by identifying mean differences between groups of patients and healthy controls. However, combining data from different subjects and groups to be able to make meaningful inferences is not trivial. The structure of the brain in each individual is unique in the size and shape as well as in the relative location of anatomical landmarks (e.g. sulci and gyri). When looking for mean differences in functional images, this issue is exacerbated by the presence of variability in functional localization i.e. variability in the location of functional regions in the brain. This is notably an important reason to focus on looking for inter-individual differences or variability.


Inter-subject variability in neuroimaging experiments is often viewed as noise. The analyses are setup in a manner to ignore this variability assuming that a global spatial normalization brings the data into the same space. Nonetheless, functional activation patterns can be impacted by variability in multiple ways for e.g., there could be spatial variability of the maps or variability in the spectral composition of the timecourses or variability in the connectivity between the activation patterns identified. The overarching problem this thesis seeks to contribute to, is seeking improved measures to quantify spatial, spectral and connectivity based variability and to identify associated cognitive or behavioral differences in the distribution of brain networks. We have successfully shown that different (spatial and spectral) measures of variability in blind source separated functional activation patterns underline previously unexplained characteristics that help in discerning schizophrenia patients from healthy controls. Additionally, we show that variance measures in dynamic connectivity between networks in healthy controls can justify relationship between connectivity patterns and executive functioning abilities.


November 18, 2015 at 10:00am - Ph.D. Imaging Science Thesis Defense - JAVIER A. CONCHA - The Use of Landsat 8 for Monitoring of Fresh and Coastal Waters

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
November 18, 2015 at 10:00am
The Use of Landsat 8 for Monitoring of Fresh and Coastal Waters
Ph.D. Imaging Science Thesis Defense



The most interaction between humankind and water occurs in coastal and inland waters at a scale of tens or hundred of meters, but there is not yet an ocean color product at this spatial scale. Landsat 8 could potentially addresses the remote sensing of these kinds of waters due to its improved features. This work presents an approach to obtain the color producing agents (CPAs) chlorophyll-a, colored dissolved organic material (CDOM) and minerals from water bodies using Landsat 8. Adequate atmospheric correction becomes an important first step to accurately retrieving water parameters since the sensor-reaching signal due to water is very small when compared to the signal due to the atmospheric effects. We developed the model-based empirical line method (MoB-ELM) atmospheric correction method. The Mob-ELM employs pseudo invariant feature (PIF) pixels extracted from a reflectance product along with the in-water radiative transfer model HydroLight. We used a look-up-table-based (LUT-based) inversion methodology to simultaneously retrieve CPAs. The LUT of remote-sensing reflectance spectra was created in Hydrolight using inherent optical properties (IOPs) measured in the field.

The retrieval algorithm was applied over three Landsat 8 scenes. The CPA concentration maps exhibit expected trends of low concentrations in clear waters and higher concentrations in turbid waters. We estimated a normalized root mean squared error (NRMSE) of about 10% for Chlorophyll-a and total suspended solid, and about 5% for colored dissolved organic matter (CDOM) when compared with in situ data. These results demonstrate that the developed algorithm allows the simultaneous mapping of concentration of all CPAs in Case 2 waters and over areas where the standard algorithms are not available due to spatial resolution. Therefore, this study shows that the Landsat 8 satellite can be utilized over Case 2 waters as long as a careful atmospheric correction is applied and IOPs are known.

November 12, 2015 at 2:00am - PHD Dissertation Defense - BIKASH BASNET - Monitoring Cloud Prone Complex Landscape At multiple Spatial Scale Using Medium and High Resolution Optical Data: A Case Study in Central Africa

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
November 12, 2015 at 2:00am
Monitoring Cloud Prone Complex Landscape At multiple Spatial Scale Using Medium and High Resolution Optical Data: A Case Study in Central Africa
PHD Dissertation Defense



Tracking land surface dynamics over cloud prone areas with complex mountainous terrain and a landscape that is heterogeneous at a scale of approximately 10m is an important challenge in the remote sensing of tropical regions in developing nations. Persistent monitoring of natural resources in these regions at multiple spatial scales requires development of tools to identify emerging land cover degradation due to anthropogenic causes such as agricultural expansion and climate change. Along with the cloud cover and obstructions by topographic distortions due to steep terrain, there are limitations to the accuracy of monitoring change using available historical satellite imagery, largely due to sparse data access and lack of high quality ground truth for classifier training.

This work addressed these problems to create an effective process for monitoring the Lake Kivu region located in Central Africa. The Lake Kivu region is a biodiversity hotspot with a complex, heterogeneous landscape and intensive agricultural development where individual plot sizes are often on the scale of 10 m. Procedures were developed that use optical data from satellite and aerial observations at multiple scales to tackle the monitoring challenges. First, a novel processing chain was developed to systematically monitor the spatio-temporal land use/land cover dynamics of this region over the years 1988, 2001, and 2011 using Landsat data, complemented by ancillary data. Topographic compensation was performed on Landsat reflectances to avoid the strong illumination angle impacts and image compositing was used to compensate for frequent cloud cover and thus incomplete annual data availability in the archive. A systematic supervised classification using state of the art machine learning classifier Random Forest was applied to the composite Landsat imagery to obtain land cover thematic maps with overall accuracies of 90% and higher. Subsequent change analysis between these years found extensive conversions of the natural environment as a result of human related activities. The gross forest cover loss for 1988-2001 and 2001- 2011 period was 216.4 and 130.5 thousand hectares, respectively, signifying significant deforestation in the period of civil war and a relatively stable and lower deforestation rate later, possibly due to conservation and reforestation efforts in the region. The other dominant land cover changes in the region were aggressive subsistence farming and urban expansion displacing natural vegetation and arable lands. Despite limited data availability, this study fills the gap of much needed detailed and updated land cover change information for this biologically important region of Central Africa.

While useful on a regional scale, Landsat data can be inadequate for more detailed studies of land cover change. Based on an increasing availability of high resolution imagery and LIDAR data from manned and unmanned aerial platforms (<1m resolution), a study was performed leading to a generic framework for land cover monitoring at fine spatial scales. The approach fuses high resolution aerial imagery and LIDAR data to produce land cover maps with high spatial detail using object-based image analysis techniques. The classification framework was tested for a scene with both natural and cultural features and found to be more than 90 percent accurate, sufficient for detailed land cover change study.



August 6, 2015 at 10:00am - M.S. Thesis Defense - Sean Archer - Empirical Measurement and Model Validation of Infrared Spectra of Liquid- Contaminated Surfaces

Carlson Bldg. (76) – Room 3215 (DIRS Lab)
August 6, 2015 at 10:00am
Sean Archer
Empirical Measurement and Model Validation of Infrared Spectra of Liquid- Contaminated Surfaces
M.S. Thesis Defense
Abstract Liquid contaminated surfaces generally require more sophisticated radiometric modeling to numerically describe surface properties. The goal of this thesis was to validate predicted infrared spectra of liquid contaminated surfaces from a recently developed micro-scale bi-directional reflectance distribution function (BRDF) model, known as microDIRSIG. This micro-scale model had been developed coincide with the Digitial Image and Remote Sensing Image Generation (DIRSIG) model as a rigorous ray tracing physics-based model capable of predicting the BRDF of geometric surfaces that are defined at micron to millimeter spatial resolution. The model offers an extension from conventional BRDF models by allowing contaminants to be added as geometric objects to a micro-facet surface. This model was validated through the use of empirical measurements. A total of 18 different substrate and contaminant combinations were measured and compared against modeled outputs. These substrates included wood and aluminum samples with three different paint finishes and varying levels of silicon based oil (SF96) liquid contamination. The longwave infrared radiance for each substrate was measured with a Design & Prototypes (D&P) Fourier transform infrared spectrometer and a Physical Sciences Inc. Adaptive Infrared Imaging Spectroradiometer (AIRIS). The microDIRSIG outputs were compared against measurements qualitatively in both the emissivity and radiance domains. A temperature emissivity separation (TES) algorithm was applied to the measured radiance spectra for comparison with the microDIRSIG predicted emissivity spectra. The model predicted emissivity spectra was also forward modeled through a DIRSIG simulation for comparisons to the measured radiance spectra. The results showed a promising agreement for homogenous surfaces with a liquid contamination that could be well characterized geometrically. Limitations arose in substrates that were modeled as homogeneous surfaces, but had spatially varying artifacts due to uncertainties with the contaminant and surface interaction. There is high desire for accurate physics based modeling of liquid contaminated surfaces and this validation framework may be extended to include a wider array of samples for more realistic natural surfaces that are often found in the real world.

August 6, 2015 at 2:00am - MS Thesis Defense - MATTHEW EDWARD MURPHY - Statistical Study of Interplanetary Coronal Mass Ejections with Strong Magnetic Fields

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
August 6, 2015 at 2:00am
Statistical Study of Interplanetary Coronal Mass Ejections with Strong Magnetic Fields
MS Thesis Defense
Abstract Coronal Mass Ejections (CMEs) with strong magnetic fields are typically associated with significant solar energetic particle (SEP) events, high solar wind speed and solar flare events. Successful prediction of the arrival time of a CME at Earth is required to maximize the time available for satellite, infrastructure, and space travel programs to take protective action against the coming flux of high-energy particles. It is known that the magnetic field strength of a CME is linked to the strength of a geomagnetic storm on Earth. Unfortunately, the correlations between strong magnetic field CMEs from the entire sun (especially from the far side or non-Earth facing side of the sun) to SEP and flare events, solar source regions and other relevant solar variables are not well known. New correlation studies using an artificial intelligence engine (Eureqa) were performed to study CME events with magnetic field strength readings over 30 nanoteslas (nT) from January 2010 to October 17, 2014. This thesis presents the results of this study, validates Eureqa to obtain previously published results, and points the way towards future studies that might extend the lead time before such events strike valuable targets.

July 28, 2015 at 2:00am - Ph.D. Imaging Science Thesis Defense - THOMAS B KINSMAN - Semi-Supervised Pattern Recognition and Machine Learning for Eye-Tracking

Carlson Fishbowl 76-1275
July 28, 2015 at 2:00am
Semi-Supervised Pattern Recognition and Machine Learning for Eye-Tracking
Ph.D. Imaging Science Thesis Defense
Abstract The first step in monitoring an observer’s eye gaze is identifying and locating the image of their pupils in video recordings of their eyes. Current systems work under a range of conditions, but fail in bright sunlight and rapidly varying illumination. A computer vision system was developed to assist with the recognition of the pupil in every frame of a video, in spite of the presence of strong first-surface reflections off of the cornea. A modified Hough Circle detector was developed that incorporates knowledge that the pupil is darker than the surrounding iris of the eye, and is able to detect imperfect circles, partial circles, and ellipses. As part of the processing, the image is modified to compensate for the distortion of the pupil caused by the out-of-plane rotation of the eye. A sophisticated noise cleaning technique was developed to mitigate first surface reflections, enhance edge contrast, and reduce image flare. Semi-supervised human input and validation is used to train the algorithm. The final results are comparable to those achieved using a human analyst, but require only a tenth of the human interaction.

July 27, 2015 at 9:00am - Ph.D. Dissertation Defense - Paul Romanczyk - Extraction of Vegetation Biophysical Structure from Small-Footprint Full-Waveform Lidar Signals

Carlson Bldg. (76) - Room 2215
July 27, 2015 at 9:00am
Paul Romanczyk
Extraction of Vegetation Biophysical Structure from Small-Footprint Full-Waveform Lidar Signals
Ph.D. Dissertation Defense


The National Ecological Observatory Network (NEON) is a continental scale environmental monitoring initiative tasked with characterizing and understanding ecological phenomenology over a 30-year time frame. To support this mission, NEON collects ground truth measurements, such as organism counts and characterization, carbon flux measurements, etc. To spatially upscale these plot-based measurements, NEON developed an airborne observation platform (AOP), with a high-resolution visible camera, next-generation AVIRIS imaging spectrometer, and a discrete and waveform digitizing light detection and ranging (lidar) system. While visible imaging, imaging spectroscopy, and discrete lidar are relatively mature technologies, our understanding of and associated algorithm development for small-footprint full-waveform lidar are still in early stages of development. This work has as its primary aim to extend small-footprint full-waveform lidar capabilities to assess vegetation biophysical structure.

In order to fully exploit waveform lidar capabilities, high fidelity geometric and radio-metric truth data are needed. Forests are structurally and spectrally complex, which makes collecting the necessary truth challenging, if not impossible. We utilize the Digital Imaging and Remote Sensing Image Generation (DIRSIG) model, which provides an environment for radiometric simulations, in order to simulate waveform lidar signals. The first step of this research was to build a virtual forest stand based on Harvard Forest inventory data. This scene was used to assess the level of geometric fidelity necessary for small-footprint waveform lidar simulation in broadleaf forests. It was found that leaves have the largest influence on the backscattered signal and that there is little contribution to the signal from the leaf stems and twigs. From this knowledge, a number of additional realistic and abstract virtual “forest” scenes were created to aid studies assessing the ability of waveform lidar systems to extract biophysical phenomenology. We developed an additive model, based on these scenes, for correcting the attenuation in backscattered signal caused by the canopy. The attenuation-corrected waveform, when coupled with estimates of the leaf-level reflectance, provides a measure of the complex within-canopy forest structure. This work has implications for our improved understanding of complex waveform lidar signals in forest environments and, very importantly, takes the research community a significant step closer to assessing fine-scale horizontally- and vertically-explicit leaf area, a holy grail of forest ecology. 

July 23, 2015 at 9:00am - Ph.D. Dissertation Defense - Rey Jan D. Garma - Image Quality Modeling and Characterization of Nyquist Sampled Framing Systems with Operational Considerations for Remote Sensing

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
July 23, 2015 at 9:00am
Rey Jan D. Garma
Image Quality Modeling and Characterization of Nyquist Sampled Framing Systems with Operational Considerations for Remote Sensing
Ph.D. Dissertation Defense
Abstract The trade between detector and optics performance is often conveyed through the Q metric, which is defined as the ratio between detector sampling frequency and optical cutoff frequency. Historically sensors have operated at Q≈1, which introduces aliasing but increases the system modulation transfer function (MTF) and signal-to-noise ratio (SNR). Though mathematically suboptimal, such designs have been operationally ideal when considering system parameters such as pointing stability and detector performance. Substantial advances in read noise and quantum efficiency of modern detectors may compensate for the negative aspects associated with balancing detector/optics performance, presenting an opportunity to revisit the potential for implementing Nyquist-sampled (Q≈2) sensors. A digital image chain simulation is developed and validated against a laboratory testbed using objective and subjective assessments. Objective assessments are accomplished by comparing the modeled MTF to measurements from slant-edge photographs. Subjective assessments are carried out by performing a psychophysical study where subjects are asked to rate simulation and testbed imagery against a ΔNIIRS scale with the aid of a marker-set. Using the validated model, additional test cases are simulated to study the effects of increased detector sampling on image quality with operational considerations. First, a factorial experiment using Q-sampling, pointing stability, integration time, and detector performance is conducted to measure the main effects and interactions of each on the response variable, ΔNIIRS. To assess the fidelity of current models, variants of the General Image Quality Equation (GIQE) are evaluated against subject-provided ratings and two modified GIQE versions are proposed. Finally, using the validated simulation and modified IQE, trades are conducted to ascertain the feasibility of implementing Q≈2 designs in future systems.

July 22, 2015 at 10:00am - Ph.D. Dissertation Defense - Tyler D. Carson - Signature Simulation and Characterization of Mixed Solids in the Visible and Thermal Regimes

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
July 22, 2015 at 10:00am
Tyler D. Carson
Signature Simulation and Characterization of Mixed Solids in the Visible and Thermal Regimes
Ph.D. Dissertation Defense
Abstract Solid target signatures vary due to geometry, chemical composition and scene radiome- try. Although radiative transfer models and function-fit physical models may describe certain targets in limited depth, the ability to incorporate all three of these signature variables is dicult. This work describes a method to simulate the transient signatures of mixed solids and soils by first considering scene geometry that was synthetically created using 3-d physics engines. Through the assignment of spectral data from the Nonconven- tional Exploitation Factors Data System (NEFDS) and other libraries, synthetic scenes are represented as a chemical mixture of particles. Finally, first principles radiometry is modeled using the Digital Imaging and Remote Sensing Image Generation (DIRSIG) model. With DIRSIG, radiometric and sensing conditions were systematically manipu- lated to produce goniometric signatures. The implementation of this virtual goniometer allows users to examine how a target bidirectional reflectance function (BRDF) and di- rectional emissivity will change with geometry, composition and illumination direction. The tool described provides geometry flexibility that is unmatched by radiative transfer models. It delivers a discrete method to avoid the significant cost of time and treasure associated with hardware based goniometric data collections.

July 14, 2015 at 2:00am - Ph.D. Dissertation Defense - Oesa A. Weaver - An Analytical Framework for Assessing the Efficacy of Small Satellites in Performing Novel Imaging Missions

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
July 14, 2015 at 2:00am
Oesa A. Weaver
An Analytical Framework for Assessing the Efficacy of Small Satellites in Performing Novel Imaging Missions
Ph.D. Dissertation Defense
Abstract In the last two decades, small satellites have opened up the use of space to groups other than governments and large corporations, allowing for increased participation and experimentation. This democratization of space was enabled by improved technology, which allowed the miniaturization of components and reduction of overall cost, meaning many of the capabilities of larger satellites could be replicated at a fraction of the cost. The potential of these smaller satellites to replace or augment existing systems has led to an explosion of potential satellite and mission concepts, often with little rigorous study of whether the proposed satellite or mission is achievable or necessary. This work proposes an analytical framework to aid system designers in evaluating the ability of an existing concept or small satellite to perform a particular imaging mission, either replacing or augmenting existing capabilities. This framework was developed and then refined by application to the problem of using small satellites to perform a wide area search mission – a mission not possible with existing imaging satellites, but one that would add to current capabilities. Requirements for a wide area search mission were developed, along with a list of factors that would affect image quality and mission performance. Two existing small satellite concepts were evaluated for use by examining image quality from the systems, selecting an algorithm to perform the search function, and then assessing mission feasibility by applying the algorithm to simulated imagery. Finally, a notional constellation design was developed to assess the number of satellites required to perform the mission. It was found that a constellation of 480 CubeSats producing 4 m spatial resolution panchromatic imagery and employing an on-board processing algorithm would be sufficient to perform a wide area search mission.

June 26, 2015 at 10:00am - Ph.D. Dissertation Defense - David Kelbe - Forest structure from terrestrial laser scanning – in support of remote sensing

Carlson Bldg. 76 - Room 3215 (DIRS Lab)
June 26, 2015 at 10:00am
David Kelbe
Forest structure from terrestrial laser scanning – in support of remote sensing
Ph.D. Dissertation Defense
Advisor: Dr. Jan van Aardt
Abstract Forests are an important part of the natural ecosystem, providing resources such as timber and fuel, performing services such as energy exchange and carbon storage, and presenting risks, such as fire damage and invasive species impacts. Improved characterization of forest structural attributes is desirable, as it could improve our understanding and management of these natural resources. Traditionally, the systematic collection of forest information related to stem volume and biomass – dubbed ``forest inventory'' - is achieved via relatively crude, readily-measured explanatory variables, such as tree height and stem diameter. Such field inventories are time-consuming, expensive, and coarse when compared to novel 3D measurement technologies. Remote sensing estimates, on the other hand, provide synoptic coverage, but often fail to capture the fine-scale structural variation of the forest environment. Terrestrial laser scanning (TLS) has demonstrated a potential to address these limitations, while offering opportunity to support remote sensing efforts by providing spatially explicit ground-truth data for calibration/validation in forest environments. An additional benefit is the potential to extract realistic 3D forest models, for use in simulation and visualization studies. However, despite this potential, operational use has remained limited due to unsatisfactory performance characteristics vs. budgetary constraints of many end-users. To address this gap, my dissertation advanced affordable mobile laser scanning capabilities for operational forest structure assessment. We developed geometric reconstruction of forest structure from rapid-scan, low-resolution point cloud data, providing for automatic extraction of standard forest inventory metrics. To augment these results over larger areas, we designed a view-invariant feature descriptor to enable marker-free registration of TLS data pairs, without knowledge of the initial sensor pose. A graph-theory framework was then integrated to perform multi-view registration between a network of disconnected scans. This provided improved structural assessment at the plot-level. Finally, a data mining approach was taken to assess plotlevel canopy structure, which has important implications to our understanding forest function. Outputs are being utilized to provide antecedent science data for NASA's HyspIRI mission and to support the National Ecological Observatory Network's (NEON) long-term environmental monitoring initiatives.

May 21, 2015 at 8:00am - Ph.D. Dissertation Defense - Katie N. Salvaggio - A Voxel-Based Approach for Imaging Voids in Three-Dimensional Point Clouds

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
May 21, 2015 at 8:00am
Katie N. Salvaggio
A Voxel-Based Approach for Imaging Voids in Three-Dimensional Point Clouds
Ph.D. Dissertation Defense


Geographically accurate scene models have enormous potential beyond that of just simple visualizations in regard to automated scene generation.  In recent years, thanks to ever increasing computational efficiencies, there has been significant growth in both the computer vision and photogrammetry communities pertaining to automatic scene reconstruction from multi-view imagery.  The result of these algorithms is a three-dimensional (3D) point cloud which can be used to obtain a final model using surface reconstruction techniques.  However, the fidelity of these point clouds has not been well studied, and voids often exist within the point cloud.  Voids exist in texturally flat areas that fail to generate features, as well as areas where multiple views were not obtained during collection, constant occlusion existed due to collection angles or overlapping scene geometry, or in regions that failed to triangulate accurately.  It may be possible to fill in small voids in the scene using surface reconstruction or hole-filling techniques, but this is not the case with larger voids, and attempting to reconstruct them using only the knowledge of the incomplete point cloud is neither accurate nor aesthetically pleasing.  

A method is presented for identifying voids in point clouds using a voxel-based approach to partition the 3D space.  By using collection geometry and information derived from the point cloud, it is possible to detect unsampled voxels such that voids can be identified.  This analysis takes into account the location of the camera and the 3D points themselves to capitalize on the idea of free space, such that voxels that lie on the ray between the camera and point are devoid of obstruction, as a clear line of sight is a necessary requirement for reconstruction.  Using this approach, voxels are classified into three categories: occupied (contains points from the point cloud), free (rays from the camera to the point passed through the voxel), and unsampled (does not contain points and no rays passed through the area).  Voids in the voxel space are manifested as unsampled voxels.  A similar line-of-sight analysis can then be used to pinpoint locations at aircraft altitude at which the voids in the point clouds could theoretically be imaged.  This work is based on the assumption that inclusion of more images of the void areas in the 3D reconstruction process will reduce the number of voids in the point cloud that were a result of lack of coverage.  Voids resulting from texturally flat areas will not benefit from more imagery in the reconstruction process, and thus are identified and removed prior to the determination of future potential imaging locations.


April 30, 2015 at 9:00am - Imaging Science MS Thesis Defense - DENGYU LIU - Efficient Space-Time Sampling with Pixel-wise Coded Exposure for High Speed Imaging

Carlson Fishbowl 76-1275
April 30, 2015 at 9:00am
Efficient Space-Time Sampling with Pixel-wise Coded Exposure for High Speed Imaging
Imaging Science MS Thesis Defense



Cameras face a fundamental tradeoff between spatial and temporal resolution. Digital still cameras can capture images with high spatial resolution, but most high-speed video cameras have relatively low spatial resolution. It is hard to overcome this tradeoff without incurring a significant increase in hardware costs. In this paper, we propose techniques for sampling, representing and reconstructing the space-time volume in order to overcome this tradeoff. Our approach has two important distinctions compared to previous works: (1) we achieve sparse representation of videos by learning an over-complete dictionary on video patches, and (2) we adhere to practical hardware constraints on sampling schemes imposed by architectures of current image sensors, which means that our sampling function can be implemented on CMOS image sensors with modified control units in the future. We evaluate components of our approach - sampling function and sparse representation by comparing them to several existing approaches. We also implement a prototype imaging system with pixel-wise coded exposure control using a Liquid Crystal on Silicon (LCoS) device. System characteristics such as field of view, Modulation Transfer Function (MTF) are evaluated for our imaging system. Both simulations and experiments on a wide range of scenes show that our method can effectively reconstruct a video from a single coded image while maintaining high spatial



April 27, 2015 at 8:00am - Ph.D. Dissertation Defense - Madhurima Bandyopadhyay - Quantifying the urban forest environment using dense discrete return LiDAR and aerial color imagery for segmentation and object-level assessment

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
April 27, 2015 at 8:00am
Madhurima Bandyopadhyay
Quantifying the urban forest environment using dense discrete return LiDAR and aerial color imagery for segmentation and object-level assessment
Ph.D. Dissertation Defense
The urban forest is becoming increasingly important in the contexts of urban green space, carbon sequestration and offsets, and socio-economic impacts. In addition to aesthetic value, these green spaces remove airborne pollutants, preserve natural resources, mitigate adverse climate changes, among other benefits. A great deal of attention recently has been paid to urban forest management. However, the comprehensive monitoring of urban vegetation for carbon sequestration and storage is an under-explored research area. Often such assessment requires information at the individual tree level, necessitating the proper masking of vegetation from the built environment, as well as delineation of individual tree crowns. As an alternative to expensive and time-consuming manual surveys, remote sensing can be used effectively in characterizing the urban vegetation and man-made objects. 
Many studies in this field has made use of aerial and multispectral/hyperspectral imagery over cities. The emergence of light detection and ranging (LiDAR) technology has provided new impetus to the effort of extracting objects and characterizing their 3D attributes - LiDAR has been used successfully to model buildings and urban trees. However, challenges remain when using such structural information only, and researchers have investigated the use of fusion-based approaches that combine LiDAR and aerial imagery to extract objects, so that the complementary characteristics of the two modalities can be utilized. 
In this study, a fusion-based classification method was implemented between aerial color (RGB) imagery and co-registered LiDAR point clouds to classify urban vegetation and buildings from other urban classes/cover types. Structural, as well as spectral features, were used in the classification method, including height, flatness, and the distribution of normal vectors from LiDAR data, along with a non-calibrated LiDAR-based vegetation index derived from combining LiDAR intensity at 1064 nm with the red channel from the RGB imagery. This novel index was dubbed the LiDAR-infused difference vegetation index (LDVI). Classification results indicated good separation between buildings and vegetation, with an overall kappa coefficient of 85%.  
A multi-tiered delineation algorithm was designed to extract individual tree crowns from the tree clusters and species-independent biomass models were developed using LiDAR-derived tree attributes in regression analysis. These LIDAR-based biomass assessments were conducted for individual trees, as well as for clusters of trees, in cases where proper delineation of individual trees was impossible. The LIDAR-derived biomass estimates were validated against allometry-based biomass estimates that were computed from field-measured tree data. The best biomass model for the tree clusters and the individual trees showed an adjusted R2 value of 0.93 and 0.58, respectively. 
The results of this study showed that the fusion-based classification approach using LiDAR and aerial color (RGB) imagery is capable of producing good object detection accuracy. It was concluded that the LDVI can be used in vegetation detection and can act as a substitute for the normalized difference vegetation index (NDVI), where multiband imagery is not available. Furthermore, the utility of LiDAR for characterizing the urban forest and associated biomass was proven. This work could have significant impact on the rapid and accurate assessment of urban green spaces and associated monitoring and management.  

April 1, 2015 at 12:00pm - Ph.D. Dissertation Defense - Amanda K. Ziemann - A manifold learning approach to target detection in high-resolution hyperspectral imagery

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
April 1, 2015 at 12:00pm
Amanda K. Ziemann
A manifold learning approach to target detection in high-resolution hyperspectral imagery
Ph.D. Dissertation Defense


Imagery collected from airborne platforms and satellites provide an important medium for remotely analyzing the content in a scene. In particular, the ability to detect a specific material within a scene is of high importance to both civilian and defense applications. This may include identifying “targets” such as vehicles, buildings, or boats. Sensors that process hyperspectral images provide the high-dimensional spectral information necessary to perform such analyses. However, research has shown that for a d-dimensional hyperspectral image, it is typical for the data to inherently occupy an m-dimensional space, with m << d. In the remote sensing community, this has led to a recent increase in the use of non-linear manifold learning, which aims to characterize the embedded lower-dimensional, non-linear manifold upon which the hyperspectral data inherently lie. Classic hyperspectral data models include statistical, linear subspace, and linear mixture models, but these can place restrictive assumptions on the distribution of the data when implementing traditional target detection approaches, and their limitations are well-documented. Here, we present an approach to target detection in HSI that is instead based on a graph theory model of the data and a manifold learning transformation, thereby avoiding these restrictive assumptions. An adaptive graph is built on the data, and then used to implement an adaptive version of locally linear embedding (LLE). We artificially induce a target manifold and incorporate it into the adaptive LLE transformation; the artificial target manifold helps to guide the separation of the target data from the background data in the new, transformed manifold coordinates. Then, target detection is performed in the manifold space. Target detection results will be shown using laboratory-measured, field-measured, and in-scene target spectra across multiple hyperspectral data sets.