<p>The main goal of this thesis is to show the strength of combining datasets from two different types of sensors to find anomalies in their data. The sensors used in this thesis are a hyperspectral camera and a scanning 3-D laser.</p><p>The report can be divided into two main parts. The first part discusses the properties of one of the datasets and how these are used to isolate anomalies. An issue to deal with here is not only what properties to look at, but how to make the process automatic. The information retained from the first dataset is then used to make intelligent choices in the second dataset. Again, one of the challenges is to make this process automatic and accurate. The second part of the project consists of presenting the results in a way that gives the most information to the user. This is done with a graphical user interface that allows the user to manipulate the way the result is presented.</p><p>The conclusion of this project is that the information from the combined sensor datasets gives better results than the sum of the information from the individual datasets. The key of success is to play to the strengths of the datasets in question. An important block of the work in this thesis, the calibration of the two sensors, was completed by Kevin Chan as his thesis work in Electrical Engineering at the University of Lund. His contribution gave access to calibrated data that supported the results presented in this report.</p>

The main goal of this thesis is to show the strength of combining datasets from two different types of sensors to find anomalies in their data. The sensors used in this thesis are a hyperspectral camera and a scanning 3-D laser. The report can be divided into two main parts. The first part discusses the properties of one of the datasets and how these are used to isolate anomalies. An issue to deal with here is not only what properties to look at, but how to make the process automatic. The information retained from the first dataset is then used to make intelligent choices in the second dataset. Again, one of the challenges is to make this process automatic and accurate. The second part of the project consists of presenting the results in a way that gives the most information to the user. This is done with a graphical user interface that allows the user to manipulate the way the result is presented. The conclusion of this project is that the information from the combined sensor datasets gives better results than the sum of the information from the individual datasets. The key of success is to play to the strengths of the datasets in question. An important block of the work in this thesis, the calibration of the two sensors, was completed by Kevin Chan as his thesis work in Electrical Engineering at the University of Lund. His contribution gave access to calibrated data that supported the results presented in this report.

<p>This dissertation explores computational optical imaging methods to circumvent the physical limitations of classical sensing. An ideal imaging system would maximize resolution in time, spectral bandwidth, three-dimensional object space, and polarization. Practically, increasing any one parameter will correspondingly decrease the others.</p><p>Spectrometers strive to measure the power spectral density of the object scene. Traditional pushbroom spectral imagers acquire high resolution spectral and spatial resolution at the expense of acquisition time. Multiplexed spectral imagers acquire spectral and spatial information at each instant of time. Using a coded aperture and dispersive element, the coded aperture snapshot spectral imagers (CASSI) here described leverage correlations between voxels in the spatial-spectral data cube to compressively sample the power spectral density with minimal loss in spatial-spectral resolution while maintaining high temporal resolution.</p><p>Photography is limited by similar physical constraints. Low f/# systems are required for high spatial resolution to circumvent diffraction limits and allow for more photon transfer to the film plain, but require larger optical volumes and more optical elements. Wide field systems similarly suffer from increasing complexity and optical volume. Incorporating a multi-scale optical system, the f/#, resolving power, optical volume and wide field of view become much less coupled. This system uses a single objective lens that images onto a curved spherical focal plane which is relayed by small micro-optics to discrete focal planes. Using this design methodology allows for gigapixel designs at low f/# that are only a few pounds and smaller than a one-foot hemisphere.</p><p>Computational imaging systems add the necessary step of forward modeling and calibration. Since the mapping from object space to image space is no longer directly readable, post-processing is required to display the required data. The CASSI system uses an undersampled measurement matrix that requires inversion while the multi-scale camera requires image stitching and compositing methods for billions of pixels in the image. Calibration methods and a testbed are demonstrated that were developed specifically for these computational imaging systems.</p> / Dissertation

<p>This dissertation addresses the problem of detecting buried explosive threats (i.e., landmines and improvised explosive devices) with ground-penetrating radar (GPR) and hyperspectral imaging (HSI) across widely-varying environmental conditions. Automated detection of buried objects with GPR and HSI is particularly difficult due to the sensitivity of sensor phenomenology to variations in local environmental conditions. Past approahces have attempted to mitigate the effects of ambient factors by designing statistical detection and classification algorithms to be invariant to such conditions. These methods have generally taken the approach of extracting features that exploit the physics of a particular sensor to provide a low-dimensional representation of the raw data for characterizing targets from non-targets. A statistical classification rule is then usually applied to the features. However, it may be difficult for feature extraction techniques to adapt to the highly nonlinear effects of near-surface environmental conditions on sensor phenomenology, as well as to re-train the classifier for use under new conditions. Furthermore, the search for an invariant set of features ignores that possibility that one approach may yield best performance under one set of terrain conditions (e.g., dry), and another might be better for another set of conditions (e.g., wet).</p><p>An alternative approach to improving detection performance is to consider exploiting differences in sensor behavior across environments rather than mitigating them, and treat changes in the background data as a possible source of supplemental information for the task of classifying targets and non-targets. This approach is referred to as context-dependent learning. </p><p>Although past researchers have proposed context-based approaches to detection and decision fusion, the definition of context used in this work differs from those used in the past. In this work, context is motivated by the physical state of the world from which an observation is made, and not from properties of the observation itself. The proposed context-dependent learning technique therefore utilized additional features that characterize soil properties from the sensor background, and a variety of nonparametric models were proposed for clustering these features into individual contexts. The number of contexts was assumed to be unknown a priori, and was learned via Bayesian inference using Dirichlet process priors.</p><p>The learned contextual information was then exploited by an ensemble on classifiers trained for classifying targets in each of the learned contexts. For GPR applications, the classifiers were trained for performing algorithm fusion For HSI applications, the classifiers were trained for performing band selection. The detection performance of all proposed methods were evaluated on data from U.S. government test sites. Performance was compared to several algorithms from the recent literature, several which have been deployed in fielded systems. Experimental results illustrate the potential for context-dependent learning to improve detection performance of GPR and HSI across varying environments.</p> / Dissertation

Microscopic imaging in the biomedical sciences allows for detailed study of the structure and function of normal and abnormal (i.e., diseased) states of cells and tissues. The expression patterns of proteins and/or physiological parameters within these specimens can be related to disease progression and prognosis, and are often heterogeneously spread throughout the entire specimen. With conventional microscopy, a large number of individual image ‘tiles’ must be captured and subsequently combined into a mosaic of the entire specimen. This has the potential to introduce artefacts at the image seams, as well as introducing non-uniform illumination of the entire specimen. A further limitation often encountered in biomedical fluorescence microscopy is the high background due to the autofluorescence (AF) of endogenous compounds within cells and tissues. Often, AF can prevent the detection and/or accurate quantification in fluorescently- labelled tissues and, in general, can reduce the reliability of results obtained from such specimens. AF spectra are relatively broad and so can be present across a large number of image spectral channels. The intensity of AF also increases as the excitation wavelength is decreased, causing increasing amounts of autofluorescence when exciting in the blue and near-UV range of the spectrum (400 - 500 nm). This thesis reports the development of hyperspectral, fluorescence and brightfield imaging of entire, paraffin-embedded, formalin-fixed (PEFF) tissue slides using a prototype confocal scanner with a large field of view (FOV). This technology addresses the challenges of imaging large tissue sections through the use of a telecentric f-theta laser scan lens thus allowing an entire microscope slide (22x70 mm) to be imaged in a single scan at resolution equivalent to a 10x microscope objective. The development and optimization of brightfield and single-channel fluorescence imaging modes are discussed in the first half of this thesis, while the second half and appendices concentrate on the spectral properties of the system and removal of AF from PEFF tissue sections. The hyperspectral imaging mode designed for this system allows the fluorescence emission spectrum of each image pixel to be sampled at 6.7 nm/channel over a spectral range of 500-700 nm. This results in the ability to separate distinct fluorescence signatures from each other, and enables quantification even in situations where the AF completely masks the signal from the applied labels.

Microscopic imaging in the biomedical sciences allows for detailed study of the structure and function of normal and abnormal (i.e., diseased) states of cells and tissues. The expression patterns of proteins and/or physiological parameters within these specimens can be related to disease progression and prognosis, and are often heterogeneously spread throughout the entire specimen. With conventional microscopy, a large number of individual image ‘tiles’ must be captured and subsequently combined into a mosaic of the entire specimen. This has the potential to introduce artefacts at the image seams, as well as introducing non-uniform illumination of the entire specimen. A further limitation often encountered in biomedical fluorescence microscopy is the high background due to the autofluorescence (AF) of endogenous compounds within cells and tissues. Often, AF can prevent the detection and/or accurate quantification in fluorescently- labelled tissues and, in general, can reduce the reliability of results obtained from such specimens. AF spectra are relatively broad and so can be present across a large number of image spectral channels. The intensity of AF also increases as the excitation wavelength is decreased, causing increasing amounts of autofluorescence when exciting in the blue and near-UV range of the spectrum (400 - 500 nm). This thesis reports the development of hyperspectral, fluorescence and brightfield imaging of entire, paraffin-embedded, formalin-fixed (PEFF) tissue slides using a prototype confocal scanner with a large field of view (FOV). This technology addresses the challenges of imaging large tissue sections through the use of a telecentric f-theta laser scan lens thus allowing an entire microscope slide (22x70 mm) to be imaged in a single scan at resolution equivalent to a 10x microscope objective. The development and optimization of brightfield and single-channel fluorescence imaging modes are discussed in the first half of this thesis, while the second half and appendices concentrate on the spectral properties of the system and removal of AF from PEFF tissue sections. The hyperspectral imaging mode designed for this system allows the fluorescence emission spectrum of each image pixel to be sampled at 6.7 nm/channel over a spectral range of 500-700 nm. This results in the ability to separate distinct fluorescence signatures from each other, and enables quantification even in situations where the AF completely masks the signal from the applied labels.

Remotely-sensed hyperspectral imagery has become one the most advanced tools for analyzing the processes that shape the Earth and other planets. Effective, rapid analysis of high-volume, high-dimensional hyperspectral image data sets demands efficient, automated techniques to identify signatures of known materials in such imagery. In this thesis, we develop a framework for automatic material identification in hyperspectral imagery using adaptive similarity measures. We frame the material identification problem as a multiclass similarity-based classification problem, where our goal is to predict material labels for unlabeled target spectra based upon their similarities to source spectra with known material labels. As differences in capture conditions affect the spectral representations of materials, we divide the material identification problem into intra-domain (i.e., source and target spectra captured under identical conditions) and inter-domain (i.e., source and target spectra captured under different conditions) settings. The first component of this thesis develops adaptive similarity measures for intra-domain settings that measure the relevance of spectral features to the given classification task using small amounts of labeled data. We propose a technique based on multiclass Linear Discriminant Analysis (LDA) that combines several distinct similarity measures into a single hybrid measure capturing the strengths of each of the individual measures. We also provide a comparative survey of techniques for low-rank Mahalanobis metric learning, and demonstrate that regularized LDA yields competitive results to the state-of-the-art, at substantially lower computational cost. The second component of this thesis shifts the focus to inter-domain settings, and proposes a multiclass domain adaptation framework that reconciles systematic differences between spectra captured under similar, but not identical, conditions. Our framework computes a similarity-based mapping that captures structured, relative relationships between classes shared between source and target domains, allowing us apply a classifier trained using labeled source spectra to classify target spectra. We demonstrate improved domain adaptation accuracy in comparison to recently-proposed multitask learning and manifold alignment techniques in several case studies involving state-of-the-art synthetic and real-world hyperspectral imagery.

The subject of this thesis is the application of Support Vector Machines on two totally different applications, facial expressions recognition and remote sensing. The basic idea of kernel algorithms is to transpose input data in a higher dimensional space, the feature space, in which linear operations on the data can be processed more easily. These operations in the feature space can be expressed in terms of input data thanks to the kernel functions. Support Vector Machines is a classifier using this kernel method by computing, in the feature space and on basis of examples of the different classes, hyperplanes that separate the classes. The hyperplanes in the feature space correspond to non linear surfaces in the input space. Concerning facial expressions, the aim is to train and test a classifier able to recognise, on basis of some pictures of faces, which emotion (among these six ones: anger, disgust, fear, joy, sad, and surprise) that is expressed by the person in the picture. In this application, each picture has to be seen has a point in an N-dimensional space where N is the number of pixels in the image. The second application is the detection of camouflage nets hidden in vegetation using a hyperspectral image taken by an aircraft. In this case the classification is computed for each pixel, represented by a vector whose elements are the different frequency bands of this pixel.

Whenever a new sensor or system comes online, engineers and analysts responsible for processing the measured data turn first to methods that are tried and true on existing systems. This is a natural, if not wholly logical approach, and is exactly what has happened in the advent of hyperspectral imagery (HSI) exploitation. However, a closer look at the assumptions made by the approaches published in the literature has not been undertaken. This thesis analyzes three key aspects of HSI exploitation: statistical data modeling, covariance estimation from training data, and dimension reduction. These items are part of standard processing schemes, and it is worthwhile to understand and quantify the impact that various assumptions for these items have on target detectability and detection statistics. First, the accuracy and applicability of the standard Gaussian (i.e., Normal) model is evaluated, and it is shown that the elliptically contoured t-distribution (EC-t) sometimes offers a better statistical model for HSI data. A finite mixture approach for EC-t is developed in which all parameters are estimated simultaneously without a priori information. Then the effects of making a poor covariance estimate are shown by including target samples in the training data. Multiple test cases with ground targets are explored. They show that the magnitude of the deleterious effect of covariance contamination on detection statistics depends on algorithm type and target signal characteristics. Next, the two most widely used dimension reduction approaches are tested. It is demonstrated that, in many cases, significant dimension reduction can be achieved with only a minor loss in detection performance. In addition, a concise development of key HSI detection algorithms is presented, and the state-of-the-art in adaptive detectors is benchmarked for land mine targets. Methods for detection and identification of airborne gases using hyperspectral imagery are discussed, and this application is highlighted as an excellent opportunity for future work.

In this thesis, a feature extraction algorithm based on 2D Local Discriminant Bases (LDB) search is developed for acoustic and hyperspectral data. The developed algorithm extracts the relevant features by both eliminating the irrelevant ones and/or by merging the ones that do not provide extra information on their own. It is implemented on real world data to separate aflatoxin contaminated or high risk hazelnuts from the sound ones by using impact acoustic and hyperspectral data. Impact acoustics data is used to sort cracked and intact shell hazelnuts with high classification accuracy. Hypespectral images of the shelled and roasted (SRT) hazelnuts are used for classification and the algorithm extracted the spectral and spatial-frequency features for that classification. Aflatoxin concentration of the SRT category hazelnuts is automatically decreased to 0.7 ppb from 608 ppb by eliminating the detected contaminated ones.