Lori Mann Bruce

Comparison of Pansharpening Algorithms: Outcome of the 2006 GRS-S Data-Fusion Contest

In January 2006, the Data Fusion Committee of the IEEE Geoscience and Remote Sensing Society launched a public contest for pansharpening algorithms, which aimed to identify the ones that perform best. Seven research groups worldwide participated in the contest, testing eight algorithms following different philosophies [component substitution, multiresolution analysis (MRA), detail injection, etc.]. Several complete data sets from two different sensors, namely, QuickBird and simulated Pleiades, were delivered to all participants. The fusion results were collected and evaluated, both visually and objectively. Quantitative results of pansharpening were possible owing to the availability of reference originals obtained either by simulating the data collected from the satellite sensor by means of higher resolution data from an airborne platform, in the case of the Pleiades data, or by first degrading all the available data to a coarser resolution and saving the original as the reference, in the case of the QuickBird data. The evaluation results were presented during the special session on data fusion at the 2006 international geoscience and remote sensing symposium in Denver, and these are discussed in further detail in this paper. Two algorithms outperform all the others, the visual analysis being confirmed by the quantitative evaluation. These two methods share the same philosophy: they basically rely on MRA and employ adaptive models for the injection of high-pass details.

Dimensionality reduction of hyperspectral data using discrete wavelet transform feature extraction

In this paper, the dyadic discrete wavelet transform is proposed for feature extraction from a high-dimensional data space. The wavelets inherent multiresolutional properties are discussed in terms related to multispectral and hyperspectral remote sensing. Furthermore, various wavelet-based features are applied to the problem of automatic classification of specific ground vegetations from hyperspectral signatures. The wavelet transform features are evaluated using an automated statistical classifier. The system is tested using hyperspectral data for various agricultural applications. The experimental results demonstrate the promising discriminant capability of the wavelet-based features. The automated classification system consistently provides over 95% and 80% classification accuracy for endmember and mixed-signature applications, respectively. When compared to conventional feature extraction methods, the wavelet transform approach is shown to significantly increase the overall classification accuracy.

Locality-Preserving Dimensionality Reduction and Classification for Hyperspectral Image Analysis

Hyperspectral imagery typically provides a wealth of information captured in a wide range of the electromagnetic spectrum for each pixel in the image; however, when used in statistical pattern-classification tasks, the resulting high-dimensional feature spaces often tend to result in ill-conditioned formulations. Popular dimensionality-reduction techniques such as principal component analysis, linear discriminant analysis, and their variants typically assume a Gaussian distribution. The quadratic maximum-likelihood classifier commonly employed for hyperspectral analysis also assumes single-Gaussian class-conditional distributions. Departing from this single-Gaussian assumption, a classification paradigm designed to exploit the rich statistical structure of the data is proposed. The proposed framework employs local Fishers discriminant analysis to reduce the dimensionality of the data while preserving its multimodal structure, while a subsequent Gaussian mixture model or support vector machine provides effective classification of the reduced-dimension multimodal data. Experimental results on several different multiple-class hyperspectral-classification tasks demonstrate that the proposed approach significantly outperforms several traditional alternatives.

Classifying mammographic mass shapes using the wavelet transform modulus-maxima method

In this article, multiresolution analysis, specifically the discrete wavelet transform modulus-maxima (mod-max) method, is utilized for the extraction of mammographic mass shape features. These shape features are used in a classification system to classify masses as round, nodular, or stellate. The multiresolution shape features are compared with traditional uniresolution shape features for their class discriminating abilities. The study involved 60 digitized mammographic images. The masses were segmented manually by radiologists, prior to introduction to the classification system. The uniresolution and multiresolution shape features were calculated using the radial distance measure of the mass boundaries. The discriminating power of the shape features were analyzed via linear discriminant analysis (LDA). The classification system utilized a simple Euclidean metric to determine class membership. The system was tested using the apparent and leave-one-out test methods. The classification system when using the multiresolution and uniresolution shape features resulted in classification rates of 83% and 80% for the apparent and leave one-out test methods, respectively. In comparison, when only the uniresolution shape features were used, the classification rates were 72 and 68% for the apparent and leave-one-out test methods, respectively.

Decision Fusion With Confidence-Based Weight Assignment for Hyperspectral Target Recognition

Conventional hyperspectral image-based automatic target recognition (ATR) systems project high-dimensional reflectance signatures onto a lower dimensional subspace using techniques such as principal components analysis (PCA), Fishers linear discriminant analysis (LDA), and stepwise LDA. Typically, these feature space projections are suboptimal. In a typical hyperspectral ATR setup, the number of training signatures (ground truth) is often less than the dimensionality of the signatures. Standard dimensionality reduction tools such as LDA and PCA cannot be applied in such situations. In this paper, we present a divide-and-conquer approach that addresses this problem for robust ATR. We partition the hyperspectral space into contiguous subspaces based on the optimization of a performance metric. We then make local classification decisions in every subspace using a multiclassifier system and employ a decision fusion system for making the final decision on the class label. In this work, we propose a metric that incorporates higher order statistical information for accurate partitioning of the hyperspectral space. We also propose an adaptive weight assignment method in the decision fusion process based on the strengths (as measured by the training accuracies) of individual classifiers that made the local decisions. The proposed methods are tested using hyperspectral data with known ground truth, such that the efficacy can be quantitatively measured in terms of target recognition accuracies. The proposed system was found to significantly outperform conventional approaches. For example, under moderate pixel mixing, the proposed approach resulted in classification accuracies around 90%, where traditional feature fusion resulted in accuracies around 65%.

Limitations of Principal Components Analysis for Hyperspectral Target Recognition

Dimensionality reduction is a necessity in most hyperspectral imaging applications. Tradeoffs exist between unsupervised statistical methods, which are typically based on principal components analysis (PCA), and supervised ones, which are often based on Fishers linear discriminant analysis (LDA), and proponents for each approach exist in the remote sensing community. Recently, a combined approach known as subspace LDA has been proposed, where PCA is employed to recondition ill-posed LDA formulations. The key idea behind this approach is to use a PCA transformation as a preprocessor to discard the null space of rank-deficient scatter matrices, so that LDA can be applied on this reconditioned space. Thus, in theory, the subspace LDA technique benefits from the advantages of both methods. In this letter, we present a theoretical analysis of the effects (often ill effects) of PCA on the discrimination power of the projected subspace. The theoretical analysis is presented from a general pattern classification perspective for two possible scenarios: (1) when PCA is used as a simple dimensionality reduction tool and (2) when it is used to recondition an ill-posed LDA formulation. We also provide experimental evidence of the ineffectiveness of both scenarios for hyperspectral target recognition applications.

Wavelet analysis of hyperspectral reflectance data for detecting pitted morningglory (Ipomoea lacunosa) in soybean (Glycine max)

This research determined the potential for wavelet-based analysis of hyperspectral reflectance signals for detecting the presence of early season pitted morningglory when intermixed with soybean and soil. Ground-level hyperspectral reflectance signals were collected in a field experiment containing plots of soybean and plots containing soybean intermixed with pitted morningglory in a conventional tillage system. The collected hyperspectral signals contained mixed reflectances for vegetation and background soil in each plot. Pure reflectance signals were also collected for pitted morningglory, soybean, and bare soil so that synthetically mixed reflectance curves could be generated, evaluated, and the mixing proportions controlled. Wavelet detail coefficients were used as features in linear discriminant analysis for automated discrimination between the soil+soybean and the soil+soybean+pitted morningglory classes. A total of 36 different mother wavelets were investigated to determine the effect of mother wavelet selection on the ability to detect the presence of pitted morningglory. When the growth stage was two to four leaves, which is still controllable with herbicide, the weed could be detected with at least 87% accuracy, regardless of mother wavelet selection. Moreover, the Daubechies 3, Daubechies 5, and Coiflet 5 mother wavelets resulted in 100% classification accuracy. Most of the best wavelet coefficients, in terms of discriminating ability, were derived from the red-edge and the nearinfrared regions of the spectrum. For comparison purposes, the raw spectral bands and principal components were evaluated as possible discriminating features. For the two-leaf to four-leaf weed growth stage, the two methods resulted in classification accuracies of 83% and 81%, respectively. The wavelet-based method was shown to be very promising in detecting the presence of early season pitted morningglory in mixed hyperspectral reflectances. D 2003 Elsevier Science Inc. All rights reserved.

Wavelets for computationally efficient hyperspectral derivative analysis

Smoothing followed by a derivative operation is often used in the analysis of hyperspectral signatures. The width of the smoothing and/or derivative operator can greatly affect the utility of the method. If one is unsure of the appropriate width or would like to conduct analysis for several widths, scale-space images can be used. This paper shows how the wavelet transform modulus-maxima method can be used to formalize and generalize the smoothing followed by derivative analysis and how the wavelet transform ran be used to greatly decrease computational costs of the analysis. The Mallat/Zhong wavelet algorithm is compared to the traditional method, convolution with Gaussian derivative filters, for computing scale-space images. Both methods are compared on two points: (1) computational expense and (2) resulting scalar decompositions. The results show that the wavelet algorithm can greatly reduce the computational expense while practically no differences exist in the subsequent scaler decompositions. The analysis is conducted on a database of hyperspectral signatures, namely, hyperspectral digital image collection experiment (HYDICE) signatures. The reduction in computational expense is by a factor of about 30, and the average Euclidean distance between resulting scale-space images is on the order of 0.02.

Why principal component analysis is not an appropriate feature extraction method for hyperspectral data

It is a popular practice in the remote sensing community to apply principal component analysis (PCA) on a high dimensional feature space to achieve dimensionality reduction. Typically, there are two primary goals for dimensionality reduction: (i) data compression and (ii) feature extraction for classification purposes. While PCA has been proven to be an optimal method for data compression, it is not necessarily an optimal method for feature extraction, particularly when the features are used in a supervised classifier. This paper addresses the issue of using PCA on hyperspectral data, specifically why PCA is not optimal for dimensionality reduction in target detection and classification applications. The authors provide theoretical and experimental analysis of PCA to demonstrate why and when PCA is not appropriate. There are variations of the Karhunen-Loeve transform that outperform PCA in a supervised classification scheme, and some of these alternative approaches are discussed in this paper.

Locality-Preserving Discriminant Analysis in Kernel-Induced Feature Spaces for Hyperspectral Image Classification

Linear discriminant analysis (LDA) has been widely applied for hyperspectral image (HSI) analysis as a popular method for feature extraction and dimensionality reduction. Linear methods such as LDA work well for unimodal Gaussian class-conditional distributions. However, when data samples between classes are nonlinearly separated in the input space, linear methods such as LDA are expected to fail. The kernel discriminant analysis (KDA) attempts to address this issue by mapping data in the input space onto a subspace such that Fishers ratio in an intermediate (higher-dimensional) kernel-induced space is maximized. In recent studies with HSI data, KDA has been shown to outperform LDA, particularly when the data distributions are non-Gaussian and multimodal, such as when pixels represent target classes severely mixed with background classes. In this letter, a modified KDA algorithm, i.e., kernel local Fisher discriminant analysis (KLFDA), is studied for HSI analysis. Unlike KDA, KLFDA imposes an additional constraint on the mapping-it ensures that neighboring points in the input space stay close-by in the projected subspace and vice versa. Classification experiments with a challenging HSI task demonstrate that this approach outperforms current state-of-the-art HSI-classification methods.