<div>Increasing the capability of maize hybrids to use nitrogen (N) more efficiently is a common goal that contributes to reducing farmer costs and limiting negative environmental impacts. However, development of such hybrids is costly and arduous due to the repeated need for laborious field and laboratory measurements of whole-plant biomass and N uptake in large early-stage breeding programs. This research evaluated alternative in-season methodologies, including field-based physiological measurements and hyperspectral remote imagery, as surrogate or predictive measures of important end-of-season N efficiency parameters. </div><div><br></div><div>Differences in the genetic potential of 285 hybrids (derived from test crosses to a single tester) with respect to N Internal Efficiency (NIE, grain yield per unit of accumulated plant N) were investigated at two Indiana locations in 2015. The hybrids (representing both early and late maturity groups) were grown at one low N rate and a single plant density. Germplasm sources included USDA, Dow AgroSciences, and “control” checks. Various secondary traits (plant height, stalk diameter, LAI, green leaf counts, and SPAD measurements) were evaluated for their potential role as surrogate measurements for N metrics at maturity (R6) such as plant N content or NIE. Four band (RGB, NIR) multispectral airborne remote sensing imagery at R1 and R3/R4 was also collected. The key findings were: 1) identification of the 10 highest and 10 lowest ranked hybrids for each maturity group in both grain yield and NIE categories, 2) the discovery of 5 top performing hybrids which had both high NIE and high yield, 3) strong correlations of leaf SPAD (at R1 and R2/R3) to grain yield or plant N at R6, 4) none of the surrogate measurements were significantly correlated to NIE, and 5) vegetation indices (NDVI and SR) from the remote imaging were not predictive of hybrid differences in yields or whole plant N content at R6. From these results we concluded genetic potential exists among current maize germplasm for NIE breeding improvements, but that more in-depth investigations were needed into other surrogate measures of relevant N efficiency traits in hybrid comparisons. </div><div><br></div><div>Next, hyperspectral imaging was investigated as a potential predictor of agronomic parameters related to N Use Efficiency (NUE, understood here as grain yield relative to applied N fertilizer input). Hyperspectral vegetation indices (HSI) were used to extract the image features for predicting N concentration (whole plant N at R6, %N), Nitrogen Conversion Efficiency (biomass per unit of plant N at R6, NCE), and NIE. Images were collected at V16/V18 and R1/R2 by manned aircraft and unmanned aerial vehicles (UAVs) at 50 cm spatial resolution. Nine maize hybrids, or a subset, were grown under N stress conditions with two plant densities over three site years in either 2014 or 2017. Forty HSI-based mixed models were analyzed for their predictability relative to the ground reference values of %N, NCE, and NIE. Two biomass and structural indices (HBSI1_682,855 and HBSI2_682,910 at R1) were predictive of NCE values and capably differentiated the highest and lowest ranked NCE hybrids. The highest prediction accuracy for NIE was achieved by two biochemical indices (HBCI_8515,550 at both V16 and R1, and HBCI9_490,550 at R1). These also allowed for hybrid differentiation of the highest and lowest ranked NIE hybrids. From these results, we concluded that accurate end-season prediction of hybrid differences in NCE and NIE was possible from mid-season hyperspectral imaging (yet not for %N). Furthermore, the quality of the predictions was dependent on image timing, actual HSI, and the targeted N parameter at maturity. </div><div><br></div><div>One benefit to hyperspectral imaging is that it can provide greater discrimination of imaged materials through its narrow, contiguous bands. However, the data are highly correlated in some ranges. This problem was mitigated through the use of partial least squares regressions (PLSR) to predict the three N parameters from the field data described previously. Data were divided into train and test; then ten-fold cross validation was performed. The twelve models evaluated included those with 89 image bands of 5 nm widths and a selected, reduced set of hyperspectral bands as predictors. The key findings were that PLSR models based on V16 and R1 images provided accurate predictions for final whole-plant %N (R² = 0.73, CV = 11%; R² = 0.68, CV = 10%) and NCE at R6 (R² = 0.71, CV = 10%; R² = 0.73, CV = 12%) but not NIE. Additionally, accurate hybrid differentiation was possible with the combination of hyperspectral imaging and PLSR at R1 to predict %N and NCE values at R6 stage. </div><div><br></div><div>The PLSR and HSI results from this research showed that hyperspectral imaging has the potential for prediction of NUE parameters through non-destructive remote sensing at a broad scale. Additional validation is needed through the study of other genotypes and locations. Nevertheless, practical application of these methods through the integration into early stage breeding programs may allow the early identification of the highest and lowest ranked hybrids providing data-driven decisions for selecting genotypes. Implementation of improved imaging approaches may drive the increased development of maize hybrids with superior NUE. </div><div><br></div>

In this dissertation, an ensemble non-uniform spectral feature selection and a kernel density decision fusion framework are proposed for the classification of hyperspectral data using a support vector machine classifier. Hyperspectral data has more number of bands and they are always highly correlated. To utilize the complete potential, a feature selection step is necessary. In an ensemble situation, there are mainly two challenges: (1) Creating diverse set of classifiers in order to achieve a higher classification accuracy when compared to a single classifier. This can either be achieved by having different classifiers or by having different subsets of features for each classifier in the ensemble. (2) Designing a robust decision fusion stage to fully utilize the decision produced by individual classifiers. This dissertation tests the efficacy of the proposed approach to classify hyperspectral data from different applications. Since these datasets have a small number of training samples with larger number of highly correlated features, conventional feature selection approaches such as random feature selection cannot utilize the variability in the correlation level between bands to achieve diverse subsets for classification. In contrast, the approach proposed in this dissertation utilizes the variability in the correlation between bands by dividing the spectrum into groups and selecting bands from each group according to its size. The intelligent decision fusion proposed in this approach uses the probability density of training classes to produce a final class label. The experimental results demonstrate the validity of the proposed framework that results in improvements in the overall, user, and producer accuracies compared to other state-of-the-art techniques. The experiments demonstrate the ability of the proposed approach to produce more diverse feature selection over conventional approaches.

In this dissertation, three studies were done for radar and hyperspectral remote sensing applications using statistical techniques. The first study investigated a relationship between synthetic aperture radar backscatter and in situ soil properties for levee monitoring. A series of statistical analyses were performed to investigate potential correlations between three independent polarization channels of radar backscatter and various soil properties. The results showed a weak but considerable correlation between the cross-polarized (HV) radar backscatter coefficients and several soil properties. The second study performed effective statistical feature extraction for levee slide classification. Images about a levee are often very large, and it is difficult to monitor levee conditions quickly because of high computational cost and large memory requirement. Therefore, a time-efficient method to monitor levee conditions is necessary. The traditional support vector machine (SVM) did not work well on original radar images with three bands, requiring extraction of discriminative features. Gray level co-occurrence matrix is a powerful method to extract textural information from grey-scale images, but it may not be practical for a big data in terms of calculation time. In this study, very efficient feature extraction methods with spatial filtering were used, including a weighted average filter and a majority filter in conjunction with a nonlinear band normalization process. Feature extraction with these filters, along with normalized bands, yielded comparable results to gray level co-occurrence matrix with a much lower computational cost. The third study focused on the case when only a small number of ground truth labels were available for hyperspectral image classification. To overcome the difficulty of not having enough training samples, a semisupervised method was proposed. The main idea was to expand ground truth using a relationship between labeled and unlabeled data. A fast self-training algorithm was developed in this study. Reliable unlabeled samples were chosen based on SVM output with majority voting or weighted majority voting, and added to labeled data to build a better SVM classifier. The results showed that majority voting and weighted majority voting could effectively select reliable unlabeled data, and weighted majority voting yielded better performance than majority voting.

The compressed sensing (CS) model of signal processing, while offering many unique advantages in terms of low-cost sensor design, poses interesting challenges for both signal acquisition and recovery, especially for signals of large size. In this work, we investigate how CS might be applied practically and efficiently in the context of natural video. We make use of a CS video acquisition approach in line with the popular single-pixel camera framework of blind, nonaptive, random sampling while proposing new approaches for the subsequent recovery of the video signal which leverage interrame redundancy to minimize recovery error. We introduce a method of approximation, which we term multihypothesis (MH) frame prediction, to create accurate frame predictions by comparing hypotheses drawn from the spatial domain of chosen reference frames to the non-overlapping, block-by-block CS measurements of subsequent frames. We accomplish this frame prediction via a novel distance-weighted Tikhonov regularization technique. We verify through our experiments that MH frame prediction via distance-weighted regularization provides state-of-the-art performance for the recovery of natural video sequences from blind CS measurements. The distance-weighted regularization we propose need not be limited to just frame prediction for CS video recovery, but may also be used in a variety of contexts where approximations must be generated from a set of hypotheses or training data. To show this, we apply our technique to supervised hyperspectral image (HSI) classification via a novel classifier we term the nearest regularized subspace (NRS) classifier. We show that the distance-weighted regularization used in the NRS method provides greater classification accuracy than state-of-the-art classifiers for supervised HSI classification tasks. We also propose two modifications to the core NRS classifier to improve its robustness to variation of input parameters and and to further increase its classification accuracy.

In this dissertation, new digital image processing methods for hyperdimensional imagery are developed and experimentally tested on remotely sensed Earth observations and medical imagery. The high dimensionality of the imagery is either inherent due to the type of measurements forming the image, as with imagery obtained with hyperspectral sensors, or the result of preprocessing and feature extraction, as with synthetic aperture radar imagery and digital mammography. In the first study, two omni-directional adaptations of gray level co-occurrence matrix analysis are developed and experimentally evaluated. The adaptations are based on a previously developed rubber band straightening transform that has been used for analysis of segmented masses in digital mammograms. The new methods are beneficial because they can be applied to imagery where the region of interest is either poorly segmented or not segmented. The methods are based on the concept of extracting circular windows s around each pixel in the image which are radially resampled to derive rectangular images. The images derived from the resampling are then suitable for standard GLCM techniques. The methods were applied to both remotely sensed synthetic aperture radar imagery, for the purpose of automated detection of landslides on earthen levees, and to digital mammograms, for the purpose of automated classification of masses as either benign or malignant. Experimental results show the newly developed methods to be valuable for texture feature extraction and classification of un-segmented objects. In the second study, a new technique of using spatial information in spectral band grouping for remotely sensed hyperspectral imagery is developed and experimentally tested. The technique involves clustering the spectral bands based on similarity of spatial features extracted from each band. The newly developed technique is evaluated in automated classification systems that utilize a single classifier and systems that utilize multiple classifiers combined with decision fusion. The systems are experimentally tested on remotely sensed imagery for agricultural applications. The spatial-spectral band grouping approach is compared to uniform band windowing and spectral only band grouping. The results show that the spatial-spectral band grouping method significantly outperforms both of the comparison methods, particularly when using multiple classifiers with decision fusion.

A process for the use of multihypothesis prediction in the reconstruction of images is proposed for use in both compressed-sensing reconstruction as well as single-image super-resolution. Specifically, for compressed-sensing reconstruction of a single still image, multiple predictions for an image block are drawn from spatially surrounding blocks within an initial non-predicted reconstruction. The predictions are used to generate a residual in the domain of the compressed-sensing random projections. This residual being typically more compressible than the original signal leads to improved compressed-sensing reconstruction quality. To appropriately weight the hypothesis predictions, a Tikhonov regularization to an ill-posed least-squares optimization is proposed. An extension of this framework is applied to the compressed-sensing reconstruction of hyperspectral imagery is also studied. Finally, the multihypothesis paradigm is employed for single-image superresolution wherein each patch of a low-resolution image is represented as a linear combination of spatially surrounding hypothesis patches.

In this dissertation, the general problem of the dimensionality reduction of hyperspectral imagery is considered. Data dimension can be reduced through compression, in which an original image is encoded into bitstream of greatly reduced size; through application of a transformation, in which a high-dimensional space is mapped into a low-dimensional space; and through a simple process of subsampling, wherein the number of pixels is reduced spatially during image acquisition. All three techniques are investigated in the course of the dissertation. For data compression, an approach to calculate an operational bitrate for JPEG2000 in conjunction with principal component analysis is proposed. It is shown that an optimal bitrate for such a lossy compression method can be estimated while maintaining both class separability as well as anomalous pixels in the original data. On the other hand, the transformation paradigm is studied for spectral dimensionality reduction; specifically, dataindependent random spectral projections are considered, while the compressive projection principal component analysis algorithm is adopted for data reconstruction. It is shown that, by incorporating both spectral and spatial partitioning of the original data, reconstruction accuracy can be improved. Additionally, a new supervised spectral dimensionality reduction approach using a sparsity-preserving graph is developed. The resulting sparse graph-based discriminant analysis is seen to yield superior classification performance at low dimensionality. Finally, for spatial dimensionality reduction, a simple spatial subsampling scheme is considered for a multitemporal hyperspectral image sequence, such that the original image is reconstructed using a sparse dictionary learned from a prior image in the sequence.

Normothermic machine perfusion (NMP) allows for ex vivo viability and functional assessment prior to liver transplantation (LT). Hyperspectral imaging represents a suitable, non-invasive method to evaluate tissue morphology and organ perfusion during NMP. Liver allografts were subjected to NMP prior to LT. Serial image acquisition of oxygen saturation levels (StO2), organ hemoglobin (THI), near-infrared perfusion (NIR) and tissue water indices (TWI) through hyperspectral imaging was performed during static cold storage, at 1h, 6h, 12h and at the end of NMP. The readouts were correlated with perfusate parameters at equivalent time points. Twentyone deceased donor livers were included in the study. Seven (33.0%) were discarded due to poor organ function during NMP. StO2 (p < 0.001), THI (p < 0.001) and NIR (p = 0.002) significantly augmented, from static cold storage (pre-NMP) to NMP end, while TWI dropped (p = 0.005) during the observational period. At 12–24h, a significantly higher hemoglobin concentration (THI) in the superficial tissue layers was seen in discarded, compared to transplanted livers (p = 0.036). Lactate values at 12h NMP correlated negatively with NIR perfusion index between 12 and 24h NMP and with the delta NIR perfusion index between 1 and 24h (rs = −0.883, p = 0.008 for both). Furthermore, NIR and TWI correlated with lactate clearance and pH. This study provides first evidence of feasibility of hyperspectral imaging as a potentially helpful contact-free organ viability assessment tool during liver NMP.

No description available.

This study presents the integration of light detection and range (LiDAR) and hyperspectral remote sensing to create a three-dimensional bird habitat map in the Greenbelt Corridor of the Elm Fork of the Trinity River. This map permits to examine the relationship between forest stand structure, landscape heterogeneity, and bird community composition. A biannual bird census was conducted at this site during the breeding seasons of 2009 and 2010. Census data combined with the three-dimensional map suggest that local breeding bird abundance, community structure, and spatial distribution patterns are highly influenced by vertical heterogeneity of vegetation surface. For local breeding birds, vertical heterogeneity of canopy surface within stands, connectivity to adjacent forest patches, largest forest patch index, and habitat (vegetation) types proved to be the most influential factors to determine bird community assemblages. Results also highlight the critical role of secondary forests to increase functional connectivity of forest patches. Overall, three-dimensional habitat descriptions derived from integrated LiDAR and hyperspectral data serve as a powerful bird conservation tool that shows how the distribution of bird species relates to forest composition and structure at various scales.