In the past three years, there has been significant interest in hyperspectral imagery (HSI) classification using vision Transformers for the analysis of remotely sensed data. Previous research predominantly focused on the empirical integration of convolutional neural networks (CNNs) to augment the network’s capability to extract local feature information. Yet, the theoretical justification for vision Transformers out-performing CNN architectures in HSI classification remains a question. To address this issue, a unified hierarchical spectral vision Transformer architecture, specifically tailored for HSI classification, is investigated. In this streamlined yet effective vision Transformer architecture, multiple mixer modules are strategically integrated separately. These include the CNN mixer, which executes convolutional operations; the spatial self-attention (SSA) mixer and channel self-attention (CSA) mixer, both of which are adaptations of classical self-attention blocks; and hybrid models, such as the SSA + CNN mixer and CSA + CNN mixer, which merge convolution with self-attention operations. This integration facilitates the development of a broad spectrum of vision Transformer-based models tailored for HSI classification. In terms of the training process, a comprehensive analysis is performed, contrasting classical CNN models and vision Transformer-based counterparts, with particular attention to disturbance robustness and the distribution of the largest eigenvalue of the Hessian. From the evaluations conducted on various mixer models rooted in the unified architecture, it is concluded that the unique strength of vision Transformers can be attributed to their overarching architecture, rather than being exclusively reliant on individual multihead self-attention (MSA) components. Extensive experiments demonstrate that the derived vision Transformer models, based on the unified architecture, surpass the classical methods when applied to multiple hyperspectral benchmark datasets.

Hyperspectral image (HSI) classification using vision Transformers is an area of active research in the field of remote sensing. However, vision Transformers suffer from lack of inductive bias inherent in convolutional neural networks (CNNs), which hinders their generalization ability when working with limited amounts of data. Furthermore, some research indicates that the competitive performance of vision Transformers primarily stems from the overall architecture, characterized by alternating residual layers of sequence mixers and multi-layer perceptron (MLP) blocks, rather than solely relying on the multi-head self-attention (MSA) mixer module. To this end, a novel CNN-mixer hierarchical spectral Transformer applied to hyperspectral image classification is proposed. To extract spectral features, a hierarchical spectral Transformer is designed to aggregate discriminative features from the spectral dimension. To remedy the lack of inductive bias inherent in vision Transformers, a simple CNN mixer is substituted for the MSA mixer to capture the local spatial features of each HSI window patch. Accordingly, the proposed network architecture leverages advantages of both CNNs and vision Transformers. Through extensive experiments, we find that the proposed method outperforms the state-of-the-art (SOTA) methods applied to the University of Houston, Botswana, and Salinas HSI datasets.

In geospatial image analysis applications, many factors can result in statistical differences between training and testing/deployment conditions. Domain adaptation techniques aim to reduce these disparsities with the goal of improving image analysis performance. Despite recent progress, some challenges remain, such as insufficient discriminability in the aligned space and negative transfer. We introduce a novel semi-supervised domain adaptation approach that improves the adversarial discriminative domain adaptation framework, addressing these challenges. We validate its effectiveness using two real-world hyperspectral image analysis datasets with varying acquisition conditions.

Exploring fast and effective spectral-spatial feature extraction algorithms for hyperspectral image (HSI) classification is one of the most focus problems in current hyperspectral remote-sensing research. Generally, the size of homogeneous regions in HSIs is not consistent in real scenario and real scenario usually consist of ground objects of different scales. Multiscale strategy starts to be used to construct discriminative features at different scales for HSI classification in recent years. To efficiently characterize the multiscale spectral-spatial features of HSIs, a curvature filters-based multiscale feature extraction method with multiscale superpixel segmentation constraint is proposed. The proposed algorithm is composed of the following major stages. First, global multiscale spectral-spatial features are efficiently extracted via progressively curvature filtering and downsampling operations, which can be regarded as an image pyramid decomposition method. Next, a multiscale superpixel segmentation strategy is applied on the first layer of the image pyramid, and a weighted mean operation is applied within and among superpixels to extract the local multiscale spatial features (LMSFs). Finally, the global multiscale curvature features (GMCFs) and the superpixel segmentation-based LMSFs are fused to form the final multiscale spectral-spatial features for classification purposes. To verify the capabilities of the proposed method, comprehensive experiments are performed on five real hyperspectral datasets. Experimental results demonstrate that the proposed method can significantly improve the classification accuracies compared to several standard HSI feature extraction and classification methods, especially when the number of samples for training is limited.

Flowering time (time to flower after planting) is important for estimating plant development and grain yield for many crops including sorghum. Flowering time of sorghum can be approximated by counting the number of panicles (clusters of grains on a branch) across multiple dates. Traditional manual methods for panicle counting are time-consuming and tedious. In this paper, we propose a method for estimating flowering time and rapidly counting panicles using RGB images acquired by an Unmanned Aerial Vehicle (UAV). We evaluate three different deep neural network structures for panicle counting and location. Experimental results demonstrate that our method is able to accurately detect panicles and estimate sorghum flowering time.

Multi-temporal Remote Sensing (RS) data acquired by multisensor mobile mapping systems such as Unmanned Aerial Vehicles (UAVs) provide significant utility for collecting plant phenotypic traits. Biomass is a plant trait that is highly correlated with biofuel production, yet also highly affected by genetics and the environment. Previous studies demonstrated the effectiveness of Recurrent Neural Networks (RNNs) for predicting end-of-season biomass in a single year. However, biomass prediction across multiple years remains a challenge due to the variation of environments and breeding trials. This paper focuses on exploring the transferability of an RNN model incorporating features extracted from LiDAR, hyperspectral and weather data. A k-means assisted transfer learning strategy is proposed to identify optimal samples for fine-tuning the pre-trained RNN model. Results from multiple experiments conducted on sorghum breeding trials at the Purdue University are summarized.

Deep learning applications are rapidly advancing in agriculture, and phenotyping in particular. Identifying the locations of plant centers and counting are critical for assessing stand quality. Yet, its precise measurement after the emergence of plants is impractical in large-scale production fields due to the labor required. In this paper, we propose a weakly supervised deep learning framework for detecting the centers and counting maize plants using high resolution georeferenced RGB data acquired from a UAV platform. We evaluate the performance of the proposed method over two dates. The obtained results show that the proposed method can efficiently identify the locations of plant centers, and thereby the plant counts, reducing both the manual field counting and human labeling required for image-based analysis.

Recent advances in remote sensing technology and algorithms for data analysis potentially provide significant utility for agricultural applications. This paper focuses on development of a data analytics based predictive modeling strategy that incorporates multi-temporal data acquired by multi-sensor data acquisition systems, and accommodates environmental inputs. Predictive models based on Recurrent Neural Networks (RNNs) are developed to incorporate high dimensional, multi-modal, multi-temporal input data. Remote Sensing (RS) features derived from Light Detection and Ranging (LiDAR) and hyperspectral inputs, as well as weather-related data are incorporated in RNN models to predict yield. Results from multiple experiments focused on high throughput phenotyping of sorghum for biomass predictions are provided and evaluated for agricultural test fields at Purdue University.

Combining spectral and spatial features in hyperspectral image classification is a common practice due to the improvements in classification accuracy that can be obtained by extracting information from neighboring pixels. However, the resulting high dimensionality of the input data and the typically limited number of labeled samples are two key challenges that affect the overall performance of supervised classification methods. To alleviate these two issues, we propose an adaptive multiview (MV)-based active learning (AL) approach that is different from the existing MV AL methods in two main ways: 1) to improve the view sufficiency, a spectral–spatial view generation approach is proposed by incorporating spatial features derived from the segmentation maps into each view and 2) to increase the diversity across views, a dynamic view is generated at each AL iteration by selecting important features from the predefined views. The performance of each view is further improved by applying the proposed AL algorithm in conjunction with an ensemble approach as back-end classifier, a scenario less explored in the remote sensing community than single classifier-based AL methodologies. The proposed approach is applied to three widely analyzed hyperspectral data sets [i.e., Kennedy Space Center (KSC), Indian Pine, and University of Houston (UH)], and the results demonstrate the efficacy of the proposed method compared with other state-of-the-art AL classification methods.

A new domain adaptation algorithm based on the class centroid and covariance alignment (CCCA) is proposed for classification of remote sensing images. This approach exploits both the first- and second-order statistics to describe the data distribution and aligns the data distribution between domains on a per-class basis. Since the predicted labels of target data are used to estimate the two statistics, we applied overall centroid alignment (OCA) as a coarse domain adaptation strategy to improve the estimation accuracy. In addition, the OCA coarse adaptation in conjunction with CCCA refined adaptation can also benefit by incorporation of spatial information, resulting in a Spa_OCA_CCCA approach. The proposed approach is easy to implement, and only one parameter is required in the spatial filtering step. It does not require labeled information in the target domain and can achieve labor-free classification. The experimental results using Hyperion, National Center for Airborne Laser Mapping, and Worldview-2 remote sensing images demonstrated the effectiveness of the proposed approach.

Unsupervised manifold learning has become accepted as an important tool for reducing dimensionality of a dataset by finding its meaningful low-dimensional representation lying on an unknown nonlinear subspace. Most manifold learning methods only embed an existing dataset, but do not provide an explicit mapping function for novel out-of-sample data, thereby potentially resulting in an ineffective tool for classification purposes, particularly for iterative methods such as active learning. To address this issue, out-of-sample extension methods have been introduced to generalize an existing embedding of new samples. In this work, a novel out-of-sample method is introduced by utilizing High Dimensional Model Representation (HDMR) as a nonlinear multivariate regression with the Tikhonov regularizer for unsupervised manifold learning algorithms. The proposed method was extensively analyzed using illustrative datasets sampled from known manifolds. Several experiments with 3D synthetic datasets and face recognition datasets were also conducted, and the performance of the proposed method was compared to several well-known out-of-sample methods. The results obtained with Locally Linear Embedding (LLE), Laplacian Eigenmaps (LE), and t-Distributed Stochastic Neighbor Embedding (t-SNE) showed that the proposed method achieves competitive even better performance than the other out-of-sample methods.

Due to the advances in technological and industrial fields, remote sensing has been adopted to a considerable extent in precision agricultural applications. Over the past few years, remote sensing utilized Mobile Mapping Systems (MMS) as the platforms for agricultural data collection. For accurate generation of georeferenced products using such MMSs, there should be a robust calibration approach that can accurately estimate the mounting parameters of the involved sensors, i.e., LiDAR unit, camera, and hyperspectral push-broom scanner. In this paper, we propose novel calibration approaches for various sensors onboard a UAV platform - 1) simultaneous estimation of lever arm and boresight angles relating LiDAR unit and camera to the GNSS/INS unit, and 2) estimation of boresight angles relating hyperspectral push-broom scanner and the GNSS/TNS unit.

Accurate phenotyping with unmanned aerial vehicles is a remote sensing application that has received recent attention as plant breeders seek to automate the expensive and time consuming traditional manual acquisition of measurements of plant traits. This paper focuses on the prediction of sorghum biomass utilizing high temporal and spatial resolution remote sensing data. Two methods are investigated for biomass prediction. The first uses nonlinear regression models to predict biomass directly from remote sensing data, based on features from Light Detection And Ranging (LiDAR) point clouds and hyperspectral data. The second strategy focuses on the biophysical sorghum crop model, APSIM, first, using remote sensing data to parametrize the crop model, and then simulating the biomass. Results from both approaches are provided and evaluated for an agricultural test field at the Agronomy Center for Research and Education (ACRE) at Purdue University.

High throughput phenotyping is rapidly gaining widespread popularity due to its ability to non-destructively extract plant traits, such as plant height, canopy density, leaf and plant structure, and so on. In this study, we focus on developing a UAV-based LiDAR system to acquire accurate time-series 3D point clouds for monitoring two specific plant traits - plant height and canopy cover - which are integral for enhancing crop genetic improvement to meet the needs of future generations. Furthermore, the obtained estimates are validated by comparing the results with those obtained from wheel-based LiDAR data.

Biomass estimation is fundamental for a variety of plant ecological studies. Direct measurement of aboveground biomass by clipping and sorting is destructive, time-consuming and laborious, thus reducing the ability of extensive sampling. Various plant traits, such as plant height, canopy cover, and leaf and plant structure contribute towards its biomass. In this study, we focus on exploiting wheel-based LiDAR data over an agricultural field to perform growth monitoring and canopy cover estimation, which would play a crucial role in the future to develop a non-invasive technique for biomass prediction.

Hyperspectral image classification via supervised approaches is often affected by the high dimensionality of the spectral signatures and the relative scarcity of training samples. Dimensionality reduction (DR) and active learning (AL) are two techniques that have been investigated independently to address these two problems. Considering the nonlinear property of the hyperspectral data and the necessity of applying AL adaptively, in this paper, we propose to integrate manifold and active learning into a unique framework to alleviate the aforementioned two issues simultaneously. In particular, supervised Isomap is adopted for DR for the training set, followed by an out-of-sample extension approach to project the large amount of unlabeled samples into previously learned embedding space. Finally, AL is performed in conjunction with k-nearest neighbor (kNN) classification in the embedded feature space. Experiments on a benchmark hyperspectral dataset illustrate the effectiveness of the proposed framework in terms of DR and the feature space refinement.

Unsupervised manifold learning has become accepted as an important tool for reducing dimensionality of a data set by finding its meaningful low dimensional representation lying on an unknown nonlinear subspace. Most manifold learning methods only embed an existing data set, but do not provide an explicit mapping function for novel out-of-sample data, thereby potentially resulting in an ineffective tool for classification purposes. To address this issue, out-of-sample extension methods have been introduced to generalize an existing embedding to new samples. In this work, a meta-modelling method called High Dimensional Model Representation (HDMR) is firstly implemented as a nonlinear multivariate regression for the out-of-sample problem for non-parametric unsupervised manifold learning algorithms. Several experiments show that the proposed method outperforms several state-of-the-art out-of-sample extension methods in terms of generalization to new samples for classification experiments on two remote sensing hyperspectral data sets.

High throughput plant phenotyping has gained significant interest in the plant science community due to its potential impact in advancing the use of advanced plant genetics for problems ranging from global food security to biomass-based energy crops. While traditional collection of field-based phenotypes is manual, automated remote sensing-based methods can reduce the manual requirements, expand the number of sampled points, and accelerate associations with genotypes. In this preliminary work, we use multiple types of features derived from multi-temporal UAV-based hyperspectral and RGB image data for prediction of sorghum biomass. Considering the nonlinear properties of the spectral input features, multiple layer perception (MLP) neural networks and support vector regression (SVR) are explored for predicting dry biomass. The analysis is conducted on datasets acquired during June-August 2016 over an agricultural test field at the Agronomy Center for Research and Education (ACRE) at Purdue University.

Techniques that combine multiple types of features, such as spectral and spatial features, for hyperspectral image classification can often significantly improve the classification accuracy and produce a more reliable thematic map. However, the high dimensionality of the input data and the typically limited quantity of labeled samples are two key challenges that affect classification performance of supervised methods. In order to simultaneously deal with these issues, a regularized multimetric active learning (AL) framework is proposed which consists of three main parts. First, a regularized multimetric learning approach is proposed to jointly learn distinct metrics for different types of features. The regularizer incorporates the unlabeled data based on the neighborhood relationship, which helps avoid overfitting at early stages of AL, when the quantity of training data is particularly small. Then, as AL proceeds, the regularizer is also updated through similarity propagation, thus taking advantage of informative labeled samples. Finally, multiple features are projected into a common feature space, in which a new batch-mode AL strategy combining uncertainty and diversity is utilized in conjunction with k-nearest neighbor classification to enrich the set of labeled samples. In order to evaluate the effectiveness of the proposed framework, the experiments were conducted on two benchmark hyperspectral data sets, and the results were compared to those achieved by several other state-of-the-art AL methods.

Augmenting spectral features with spatial features for hyperspectral image classification can often improve the classification accuracy. However, the resulting high dimensionality and the typically scarce quantities of labeled samples impose significant challenges for supervised techniques. To alleviate these two issues simultaneously, in this paper, a semi-supervised multi-metric learning method is proposed for feature extraction and combined with active learning (AL) into a unique framework. In particular, the proposed metric learning approach learns distinct projection matrices jointly, and each metric is assigned to one type of feature. Moreover, the proposed regularizer helps avoid overfitting by taking advantage of the unlabeled data information. Finally, different types of features are projected into a common feature space in which AL is performed in conjunction with k-nearest neighbor (kNN) classification. Experiments on two benchmark hyperspectral datasets illustrate the effectiveness of the proposed framework compared to other state-of-the-art AL classification methods.

The classification of hyperspectral and multimodal remote sensing data is affected by two key problems: the high dimensionality of the input data and the limited number of the labeled samples. In this letter, a multimetric learning approach that combines feature extraction and active learning (AL) is introduced to deal with these two issues simultaneously. In particular, distinct metrics are assigned to different types of features and then learned jointly. In this way, multiple features are projected into a common feature space, in which AL is then performed in conjunction with k- nearest neighbor classification to enrich the set of labeled samples. Experiments on two sets of remote sensing data illustrate the effectiveness of the proposed framework in terms of both classification accuracy and computational requirements.

Classification of remotely sensed hyperspectral images via supervised approaches is typically affected by high dimensionality of the spectral data and a limited number of labeled samples. Dimensionality reduction via feature extraction and active learning (AL) are two approaches that researchers have investigated independently to deal with these two problems. In this paper, we propose a new method in which the feature extraction and AL steps are combined into a unique framework. The idea is to learn and update a reduced feature space in a supervised way at each iteration of the AL process, thus taking advantage of the increasing labeled information provided by the user. In particular, the computation of the reduced feature space is based on the large-margin nearest neighbor (LMNN) metric learning principle. This strategy is applied in conjunction with k-nearest neighbor ( k-NN) classification, for which a new sample selection strategy is proposed. The methodology is validated experimentally on four benchmark hyperspectral data sets. Good improvements in terms of classification accuracy and computational time are achieved with respect to the state-of-the-art strategies that do not combine feature extraction and AL.

Multitemporal hyperspectral images provide valuable information for a wide range of applications related to supervised classification, including long-term environmental monitoring and land cover change detection. However, the required ground reference data are time-consuming and expensive to acquire, motivating researchers to investigate options for reusing limited training data for classification of other temporal images. Current studies that address high dimensionality and nonstationarity inherent in temporal hyperspectral data for classification are limited for the case where significant spectral drift exists between images. In this paper, we adapt and extend two manifold alignment (MA) methods for classification of multitemporal hyperspectral images in a common manifold space, assuming that the local geometries of two temporal spectral images are similar. The first method exploits a locally based manifold configuration of a source image (considered to be the “prior” manifold), and the second approach links local manifolds of two images using bridging pairs. In addition to exploiting manifolds estimated with spectral information for MA, we also demonstrate how spatial information can be incorporated into the MA methods. When evaluated using three Hyperion data sets, the proposed methods outperform four baseline approaches and two state-of-the-art domain adaptation methods. The advantages of the proposed MA methods are more evident when significant spectral drift exists between two temporal images. In addition to the promising classification results, the proposed methods establish a domain adaptation framework for analysis of temporal hyperspectral data based on data geometry.

Augmenting spectral features with spatial features for hyperspectral image classification has recently gained significant attention, as classification accuracy can often be improved by extracting spatial features from neighboring pixels. However, the resulting high dimensional input data, which are often difficult and expensive to obtain, require large quantities of labeled data to train a robust supervised classifier. To alleviate the “curse of dimensionality”, we propose an ensemble based active learning approach that incorporates spatial features for each feature subset (view) independently. Specifically, in each view, the spatial features are extracted from an optimum segmentation selected from the hierarchical segmentation (HSeg). The proposed approach is applied to a benchmark hyperspectral data set, and the experimental results demonstrate the efficacy of the proposed method compared to other state-of-the-art active learning classification methods.

Graph construction, which is at the heart of graph-based semisupervised learning (SSL), is investigated by using manifold learning (ML) approaches. Since each ML method can be demonstrated to correspond to a specific graph, we build the relation between ML and SSL via the graph, where ML methods are employed for graph construction. Moreover, sparsity is important for the efficiency of SSL algorithms, and therefore, local ML (LML)-method-based sparse graphs are utilized. The LML-based graphs are able to capture the local geometric properties of hyperspectral data and, thus, are beneficial for classification of data with complex geometry and multiple submanifolds. In experiments with Hyperion and AVIRIS hyperspectral data, graphs constructed by two LML methods, namely, locally linear embedding and local tangent space alignment (LTSA), performed better than several popular graph construction methods, and the highest accuracies were obtained by using graphs provided by LTSA.

In this paper, we study a novel approach to active learning in the wavelet domain for classification of Full-waveform LiDAR (FWL) data. Unlike discrete 3-dimensional points obtained from a traditional discrete return LiDAR system, FWL systems have the capability to record the entire backscat-tered signal, which contains additional information about the reflecting objects. With such LiDAR systems, the vertical structure of reflectors is effectively characterized by the shape of the return pulse. Instead of deriving simple structure and statistics-based features, such as pulse amplitude and width, skewness, and kurtosis from the FWL data, in this work, wavelet features are extracted via a Redundant Discrete Wavelet Transform (RDWT) and are then utilized for classification in a multi-view active learning (AL) framework. Additionally, we demonstrate that the proposed approach provides a noise robust framework for analysis and classification of low Signal-to-Noise (SNR) LiDAR data. Experimental results demonstrate the efficacy of the proposed wavelet-based active learning for FWL data.

Nonlinear manifold learning based spectral unmixing provides an alternative to direct nonlinear unmixing methods for accommodating nonlinearities inherent in hyperspectral data. Although manifolds can effectively capture nonlinear features in the dimensionality reduction stage of unmixing, the computational overhead is excessive for large remotely sensed data sets. Manifold approximation using a set of distinguishing points is commonly utilized to mitigate the computational burden, but selection of these landmark points is important for adequately representing the topology of the manifold. This study proposes an active landmark sampling framework for manifold learning based spectral unmixing using a small initial landmark set and a computationally efficient backbone-based strategy for constructing the manifold. The active landmark sampling strategy selects the best additional landmarks to develop a more representative manifold and to increase unmixing accuracy.

Soil erosion is one of the processes responsible for water and soil quality deterioration and is impacted by local soil and land cover conditions. One of the primary functions of land cover is to protect the soil and prevent land degradation by water and wind erosion [1]. Recent interest in biofuel energy production can compromise soil quality due to increased removal of crop residue to be used as source of biofuel feedstocks. Knowledge of the impact of human-induced changes to land cover is critical to developing ecosystem-based management approaches to address these issues.

Adapting a pre-trained classifier with labeled samples from an image for classification of another temporally related image is a common multitemporal image classification strategy. However, the adaptation is not effective when the spectral drift exhibited in temporal data is significant. Instead of iteratively redefining classifier parameters, we exploit similar data geometries of temporal data and project temporal data into a joint manifold space where similar samples are clustered. The proposed classification framework is based on aligning global temporal data manifolds. In addition to global structures, we also consider the local scale by incorporating local point relations into the alignment process. In experiments with challenging temporal hyperspectral data, the proposed framework provides favorable classification results, compared to the baseline.

Exploiting disparate features from potentially different data sources with multiple-kernel based machine learning is a promising approach for analyzing geo-spatial data. A mixture-of-kernel approach can facilitate construction of a more effective training data pool with Active Learning (AL). In addition, this could alleviate the computational burden in AL implementations. Kernel based learning requires hyperparameter tuning for model selection. Further, an optimal function is required to integrate different features or data sources appropriately in the kernel induced space. Both kernel parameters and kernel combination functions may need to be tuned at each AL learning step, which is potentially very time-consuming. In this paper, a novel multiple kernel active learning algorithm is proposed that promises enhanced classification, improved AL performance, and a mechanism for automatic selection of kernel weights in the mixture-of-kernels. We demonstrate the usefulness of the proposed framework with results for both feature fusion and sensor fusion tasks.

Endmember extraction and unmixing methods that exploit nonlinearity in hyperspectral data are receiving increased attention, but they have significant challenges. Global feature extraction methods such as isometric feature mapping have significant computational overhead, which is often addressed for the classification problem via landmark-based methods. Because landmark approaches are approximation methods, experimental results are often highly variable. We propose a new robust landmark selection method for the purpose of pixel unmixing that exploits spectral and spatial homogeneity in a local window kernel. We compare the performance of the method to several landmark selection methods in terms of reconstruction error and processing time.

Active learning, which has a strong impact on processing data prior to the classification phase, is an active research area within the machine learning community, and is now being extended for remote sensing applications. To be effective, classification must rely on the most informative pixels, while the training set should be as compact as possible. Active learning heuristics provide capability to select unlabeled data that are the “most informative” and to obtain the respective labels, contributing to both goals. Characteristics of remotely sensed image data provide both challenges and opportunities to exploit the potential advantages of active learning. We present an overview of active learning methods, then review the latest techniques proposed to cope with the problem of interactive sampling of training pixels for classification of remotely sensed data with support vector machines (SVMs). We discuss remote sensing specific approaches dealing with multisource and spatially and time-varying data, and provide examples for high-dimensional hyperspectral imagery.

Hyperspectral sensors record the reflectance from the Earth's surface over the full range of solar wavelengths with high spectral resolution. The resulting high-dimensional data contain rich information for a wide range of applications. However, for a specific application, not all the measurements are important and useful. The original feature space may not be the most effective space for representing the data. Feature mining, which includes feature generation, feature selection (FS), and feature extraction (FE), is a critical task for hyperspectral data classification. Significant research effort has focused on this issue since hyperspectral data became available in the late 1980s. The feature mining techniques which have been developed include supervised and unsupervised, parametric and nonparametric, linear and nonlinear methods, which all seek to identify the informative subspace. This paper provides an overview of both conventional and advanced feature reduction methods, with details on a few techniques that are commonly used for analysis of hyperspectral data. A general form that represents several linear and nonlinear FE methods is also presented. Experiments using two widely available hyperspectral data sets are included to illustrate selected FS and FE methods.

Spectral unmixing methods that exploit nonlinearity in hyperspectral data are promising, but face significant computational challenges. Global dimensionality reduction methods such as ISOMAP have significant computational overhea, while local methods such as Locally Linear Embedding (LLE), are computationally less demanding, but may not be robust. We propose a new landmark selection method for spectral unmixing that exploits spectral and spatial information, and embed it in LLE, resulting in a hybrid method whose structure shares characteristics with both global and local manifolds. Performance of the method is compared to that of several landmark selection methods in terms of mean of reconstruction error and corresponding variance, processing time, and visual inspection of the fully unmixed scene.

Similarity measures for classification of hyperspectral data in the manifold space are typically based on spectral characteristics. However, samples that are not spectrally separable may cause incorrectly connected graphs and result in noninformative data manifolds. Spatial relationships inherent in remote sensing images can be beneficial for constructing connectivity graphs. A spectral-spatial proximity graph utilizing both spectral characteristics and spatial homogeneity is proposed for robust manifold learning. With the proposed spectral-spatial graph, we are able to extract essential features and preserve important knowledge in a lower dimensional manifold space, where classification tasks can be performed effectively. Two hyperspectral data sets were used to validate the proposed approach. Classification results obtained by the nearest neighbor classifier demonstrate the usefulness of exploiting spectral similarity and spatial proximity for the manifold-based classification.

Determination of structural characteristics of forests at large scales is an important problem in both scientific studies and development of management practices. Light detection and ranging (LIDAR) waveform data have been demonstrated to be valuable for estimating forest structural parameters even in dense forests, although challenges inherent to the LIDAR acquisition systems must be addressed. A new approach for processing LIDAR waveform data to estimate forest structural parameters is proposed. It was applied to Laser Vegetation Imaging Sensor waveform data acquired over old-growth tropical forest in the La Selva Biological Station, Costa Rica. Linear and nonlinear feature extraction methods were utilized to derive a lower dimensional feature space from high-dimensional LIDAR waveform data. The resulting features were used to estimate mean canopy heights through multiple linear regression analysis. Experimental results obtained by the new approach were statistically comparable to estimates obtained using features extracted via traditional waveform analysis, and the proposed approach successfully discovered another meaningful lower dimensional feature space without manual interpretation.

Active learning (AL) seeks to interactively construct a smaller training data set that is the most informative and useful for the supervised classification task. Based on the multiview Adaptive Maximum Disagreement AL method, this study investigates the principles and capability of several approaches for the view generation for hyperspectral data classification, including clustering, random selection, and uniform subset slicing methods, which are then incorporated with dynamic view updating and feature space bagging strategies. Tests on Airborne Visible/Infrared Imaging Spectrometer and Hyperion hyperspectral data sets show excellent performance as compared with random sampling and the simple version support vector machine margin sampling, a state-of-the-art AL method.

In order to focus on the hard classes in a multi-class classification task, a critical class oriented query strategy is proposed, which combines the concepts of "guided learning" and "active learning". In conjunction with the SVM classifier, hard pair classes are first identified based on the instability of the classification hyperplane, whereby category level guidance for which class should be queried next is sought and then provided to the active query system. Samples with higher possibility of belonging to these classes as evaluated by the current learner are queried first. Two methods are proposed. The first method (SVM-CC) simply conducts category level query. The second method (SVM- CCMS) further incorporates the uncertainty measurement based on the idea of margin sampling, so as to directly focus on the most informative samples from the identified "trouble classes". Experiments are conducted on AVIRIS and Hyperion data. Results are compared to Random Sampling and the state-of-the-art active learning method SVM based simple margin sampling SVMMS. Superior performance is obtained, whereas hard classes are successfully identified first.

Crop residue is an important factor in determining soil structure relative to soil organic matter content, water infiltration, evaporation, and soil temperature [1]. There is also a direct impact related to production of biofuels. The use of the NDTI (Normalized Difference Tillage Index [2]) from multispectral data and the CAI (Cellulose Absorption Index [3]) from hyperspectral data are investigated as a means of calibrating indices derived from the Advanced Land Imager (ALI) on the EO-1 satellite and Landsat TM, with the goal of improving residue cover estimates over extended areas.

While spectral and temporal advantages of multitemporal hyperspectral images provide opportunities for advancing classification of time varying phenomena, significant challenges are associated with high dimensionality and nonstationary signatures. While manifold learning retains critical geometry and develops a low dimension space where class clusters are recovered, spectral changes in temporal imagery impact the fidelity of the geometric representation of class dependent data. In this paper, we investigate a manifold alignment framework that exploits prior information while exploring similar local structures. The aim is to make use of common underlying geometries of two multitemporal images and embed the resemblances in a joint data manifold for classification tasks. Promising results support the advantages of the proposed manifold alignment approach.

A multi-view based active learning method (AMDWVE) is proposed as a means to optimally construct the training set for supervised classification of hyperspectral data, thereby reducing the effort required to acquire ground reference data. The method explores the intrinsic multi-view information embedded in hyperspectral data. By adaptively and quantitatively measuring the disagreement level of different views, the learner focuses on samples with higher confusion, rapidly reducing the version space and improving the learning speed. Classification confidence of each view towards each class is also obtained in the learning process and used to compensate for view insufficiency. Experiments show excellent performance on both unlabeled and unseen data from two sets of hyperspectral image data with 10 classes acquired by AVIRIS, as compared to random sampling and the state-of-the-art SVMSIMPLE.

Anomaly detection in hyperspectral images is investigated using local tangent space alignment (LTSA) for dimensionality reduction (DR) in conjunction with a minimum distance detector. The LTSA is implemented for large images by constructing a manifold with training data and employing the out-of-sample extension for testing data. The training data that should represent all the background types are generated by the recursive hierarchical segmentation (RHSEG) algorithm and the elimination of the very small segments that may represent anomalies. Experimental results indicate that the LTSA is able to distinguish anomalies from background using a small number of features in the embedded space, and the LTSA-based detector has superior anomaly detection performance to the well-known RX and kernel RX detectors.

The 24 papers in this special issue focus on hyperspectral image and signal processing.

Localized training data typically utilized to develop a classifier may not be fully representative of class signatures over large areas but could potentially provide useful information which can be updated to reflect local conditions in other areas. An adaptive classification framework is proposed for this purpose, whereby a kernel machine is first trained with labeled data and then iteratively adapted to new data using manifold regularization. Assuming that no class labels are available for the data for which spectral drift may have occurred, resemblance associated with the clustering condition on the data manifold is used to bridge the change in spectra between the two data sets. Experiments are conducted using spatially disjoint data in EO-1 Hyperion images, and the results of the proposed framework are compared to semisupervised kernel machines.

Approaches to combine local manifold learning (LML) and thek-nearest-neighbor (kNN) classifier are investigated for hyperspectral image classification. Based on supervised LML (SLML) andkNN, a new SLML-weightedkNN (SLML-WkNN) classifier is proposed. This method is appealing as it does not require dimensionality reduction and only depends on the weights provided by the kernel function of the specific ML method. Performance of the proposed classifier is compared to that of unsupervised LML (ULML) and SLML for dimensionality reduction in conjunction with thekNN (ULML-kNN and SLML-kNN). Three LML methods, locally linear embedding (LLE), local tangent space alignment (LTSA), and Laplacian eigenmaps, are investigated with these classifiers. In experiments with Hyperion and AVIRIS hyperspectral data, the proposed SLML-WkNN performed better than ULML-kNN and SLML-kNN, and the highest accuracies were obtained using weights provided by supervised LTSA and LLE.

Agricultural crop residues have a significant role in nutrient cycling, carbon sequestration, and soil erosion. In the Midwestern United States, residue cover is strongly related to tillage practices. Estimates of residue cover have been obtained from various multispectral and hyperspectral indices. Methods based on hyperspectral data yield superior results and are more robust to variation in soils and vegetation cover, but are limited by availability of data. Approaches that utilize multispectral data require local field data to calibrate models and have limited generalization over extended areas. The goal of this study is to investigate approaches to exploit the superior discrimination capability of hyperspectral data to improve results derived from multispectral data over extended areas and reduce dependence on local field based measurements. A classification based approach which utilizes the CAI hyperspectral index to identify labeled data associated with residue cover classes corresponding to intensity of tillage, and a regression modeling approach which relates CAI and NDTI, are evaluated to estimate residue coverage in soybean and corn fields in typical agricultural lands in central Indiana, U.S.A.

Decomposing a waveform into distinct components by fitting a mixture of Gaussians impacts the ultimate interpretation of the return waveform, because the resulting parameter estimates of the Gaussian mixture directly affect the understanding of vertical structure within laser footprints. Decomposing the waveform into a mixture of Gaussians involves two related problems; (1) determining the number of Gaussian components in the waveform, and (2) estimating the parameters of each Gaussian component of the mixture. In this study, a two-stage approach is proposed and applied over three areas with different land cover characteristics. Experimental results indicate that the proposed two-stage approach typically fits Gaussian mixtures over received waveforms using fewer components, while the SSE value is smaller than that of the corresponding NASA GLA14 product.

A classifier derived from labeled samples acquired over an extended area may not perform well for a specific sub-region if the spectral signatures of classes vary across the image. However, characterizing the local effects are an ill-posed problem, particularly for hyperspectral data, since an adequate number of labeled samples is not typically available for every location. This problem is addressed using semi-supervised learning and manifold learning, which both exploit the information provided by unlabeled samples in the image. A spatially adaptive classification method that uses Laplacian regularization is proposed, with the updating scheme using a combination of labeled and unlabeled samples.

Estimation of crop residue has increasing relevance for agricultural land management related to erosion, carbon sequestration and water quality. Automated methods utilizing remotely sensed data potentially provide capability to accomplish this over large areas with greater accuracy and at lower cost than traditional "windscreen" surveys. This study investigated application of the cellulose absorption index (CAI) and constrained linear spectral mixing analysis (CLSMA) to Hyperion hyperspectral data to estimate soy and corn crop residue coverage on agricultural lands in Indiana.

Obtaining training data for land cover classification using remotely sensed data is time consuming and expensive especially for relatively inaccessible locations. Therefore, designing classifiers that use as few labeled data points as possible is highly desirable. Existing approaches typically make use of small-sample techniques and semisupervision to deal with the lack of labeled data. In this paper, we propose an active learning technique that efficiently updates existing classifiers by using fewer labeled data points than semisupervised methods. Further, unlike semisupervised methods, our proposed technique is well suited for learning or adapting classifiers when there is substantial change in the spectral signatures between labeled and unlabeled data. Thus, our active learning approach is also useful for classifying a series of spatially/temporally related images, wherein the spectral signatures vary across the images. Our interleaved semisupervised active learning method was tested on both single and spatially/temporally related hyperspectral data sets. We present empirical results that establish the superior performance of our proposed approach versus other active learning and semisupervised methods.

Hyperspectral data provide new capability for discriminating spectrally similar classes, but unfortunately such class signatures often overlap in multiple narrow bands. Thus, it is useful to incorporate reliable spatial information when possible. However, this can result in increased dimensionality of the feature vector, which is already large for hyperspectral data. Markov random field (MRF) approaches, such as iterated conditional modes (ICM), can provide evidence relative to the class of a neighbor through Gibbs' distribution, but suffer from computational requirements and curse of dimensionality issues when applied to hyperspectral data. In this paper, a new knowledge based stacking approach is presented to utilize spatial information within homogeneous regions and at class boundaries, while avoiding the curse of dimensionality. The approach learns the location of the class boundary and combines original bands with the extracted spectral information of a neighborhood to train a hierarchical support vector machine (HSVM) classifier. The new method is applied to hyperspectral data collected by the Hyperion sensor on the EO-1 satellite over the Okavango delta of Botswana. Classification accuracies are compared to those obtained by a pixel-wise HSVM classifier, majority filtering and ICM to demonstrate the advantage of the knowledge based stacking approach.

Nonlinear manifold learning algorithms assume that the original high dimensional data actually lie on a low dimensional manifold defined by local geometric distances between samples. Most of the traditional methods have focused only on the spectral distances in calculating the local dissimilarity of samples, whereas in the case of image data, the spatial distribution and localized contextual information of image samples could provide useful information. As a framework for integrating spatial and spectral information associated with image samples, a hierarchical spatial-spectral segmentation method is investigated for constructing the manifold structure. The new approach, which develops the manifold for the purpose of classification, incorporates an updating scheme whereby the spatial information and class labels are transferred through the segmentation hierarchy. It is applied to hyperspectral data collected by the Hyperion sensor on the EO-1 satellite over the Okavango Delta of Botswana. Classification accuracies and generalization capability are compared to those achieved by the best basis binary hierarchical classifier, the hierarchical support vector machine classifier, and the shortest path k-nearest neighbor classifier.

The 19 articles in this special issue focus on remote sensing for major disaster prevention, monitoring, and assessment. Topics include earthquakes and landslides, tsunami, hurricanes and typhoons, floods and fires, as well as papers with a broader focus, highlighting innovative tools and procedures to exploit Earth observation data.

Nonlinear manifold learning algorithms, mainly isometric feature mapping (Isomap) and local linear embedding (LLE), determine the low-dimensional embedding of the original high dimensional data by finding the geometric distances between samples. Researchers in the remote sensing community have successfully applied Isomap to hyperspectral data to extract useful information. Although results are promising, computational requirements of the local search process are exhorbitant. Landmark-Isomap, which utilizes randomly selected sample points to perform the search, mitigates these problems, but samples of some classes are located in spatially disjointed clusters in the embedded space. We propose an alternative approach to selecting landmark points which focuses on the boundaries of the clusters, rather than randomly selected points or cluster centers. The unique Isomap is evaluated by SStress, a good- of-fit measure, and reconstructed with reduced computation, which makes implementation with other classifiers plausible for large data sets. The new method is implemented and applied to Hyperion hyperspectral data collected over the Okavango Delta of Botswana.

Obtaining ground truth for classification of remotely sensed data is time consuming and expensive. In addition, a number of factors cause the spectral signatures of the same class to vary spatially. Therefore, successful adaptation of a classifier designed from available labeled data to classify new images acquired over other geographic locations is difficult but invaluable to the remote sensing community. In this paper we propose an active learning technique for rapidly updating existing classifiers using very few labeled data points from the new image. We also show empirically that our updated classifier exhibits better learning rates than classifiers trained via other active learning and semi-supervised methods.

Obtaining ground truth for classification of remotely sensed data is time consuming and expensive, resulting in poorly represented signatures over large areas. In addition, the spectral signatures of a given class vary with location and/or time. Therefore, successful adaptation of a classifier designed from the available labeled data to classify new hyperspectral images acquired over other geographic locations or subsequent times is difficult, if minimal additional labeled data are available. In this paper, the binary hierarchical classifier is used to propose a knowledge transfer framework that leverages the information extracted from the existing labeled data to classify spatially separate and multitemporal test data. Experimental results show that in the absence of any labeled data in the new area, the approach is better than a direct application of the original classifier on the new data. Moreover, when small amounts of the labeled data are available from the new area, the framework offers further improvements through semisupervised learning mechanisms and compares favorably with previously proposed methods

Abstract — The shortest path k-nearest neighbor classifier (SkNN), that utilizes nonlinear manifold learning, is proposed for analysis of hyperspectral data. In contrast to classifiers that deal with the high dimensional feature space directly, this approach uses the pairwise distance matrix over a nonlinear manifold to classify novel observations. Because manifold learning preserves the local pairwise distances and updates distances of a sample to samples beyond the user-defined neighborhood along the shortest path on the manifold, similar samples are moved into closer proximity. High classification accuracies are achieved by using the simple k-nearest neighbor (kNN) classifier. SkNN was applied to hyperspectral data collected by the Hyperion sensor on the EO-1 satellite over the Okavango Delta of Botswana. Classification accuracies and generalization capability are compared to those achieved by the best basis binary hierarchical classifier, the hierarchical support vector machine classifier, and the k-nearest neighbor classifier on both the original data and a subset of its principal components. I.

Statistical classification of byperspectral data is challenging because the inputs are high in dimension and represent multiple classes that are sometimes quite mixed, while the amount and quality of ground truth in the form of labeled data is typically limited. The resulting classifiers are often unstable and have poor generalization. This work investigates two approaches based on the concept of random forests of classifiers implemented within a binary hierarchical multiclassifier system, with the goal of achieving improved generalization of the classifier in analysis of hyperspectral data, particularly when the quantity of training data is limited. A new classifier is proposed that incorporates bagging of training samples and adaptive random subspace feature selection within a binary hierarchical classifier (BHC), such that the number of features that is selected at each node of the tree is dependent on the quantity of associated training data. Results are compared to a random forest implementation based on the framework of classification and regression trees. For both methods, classification results obtained from experiments on data acquired by the National Aeronautics and Space Administration (NASA) Airborne Visible/Infrared Imaging Spectrometer instrument over the Kennedy Space Center, Florida, and by Hyperion on the NASA Earth Observing 1 satellite over the Okavango Delta of Botswana are superior to those from the original best basis BHC algorithm and a random subspace extension of the BHC.

A new multistage method using hierarchical clustering for unsupervised image classification is presented. In the first phase, the multistage method performs segmentation using a hierarchical clustering procedure which confines merging to spatially adjacent clusters and generates an image partition such that no union of any neighboring segments has homogeneous intensity values. In the second phase, the segments resulting from the first stage are classified into a small number of distinct states by a sequential merging operation. The region-merging procedure in the first phase makes use of spatial contextual information by characterizing the geophysical connectedness of a digital image structure with a Markov random field, while the second phase employs a context-free similarity measure in the clustering process. The segmentation procedure of region merging is implemented as a hierarchical clustering algorithm whereby a multiwindow approach using a pyramid-like structure is employed to increase computational efficiency while maintaining spatial connectivity in merging. From experiments with both simulated and remotely sensed data, the proposed method was determined to be quite effective for unsupervised analysis. In particular, the region-merging approach based on spatial contextual information was shown to provide more accurate classification of images with smooth spatial patterns.

This paper presents a new approach called Hierarchical Support Vector Machines (HSVM), to address multiclass problems. The method solves a series of maxcut problems to hierarchically and recursively partition the set of classes into two-subsets, till pure leaf nodes that have only one class label, are obtained. The SVM is applied at each internal node to construct the discriminant function for a binary metaclass classifier. Because maxcut unsupervised decomposition uses distance measures to investigate the natural class groupings. HSVM has a fast and intuitive SVM training process that requires little tuning and yields both high accuracy levels and good generalization. The HSVM method was applied to Hyperion hyperspectral data collected over the Okavango Delta of Botswana. Classification accuracies and generalization capability are compared to those achieved by the Best Basis Binary Hierarchical Classifier, a Random Forest CART binary decision tree classifier and Binary Hierarchical Support Vector Machines.

A multistage hierarchical clustering technique, which is an unsupervised technique, has been proposed in this paper for classifying the hyperspectral data. The multistage algorithm consists of two stages. The "local" segmentor of the first stage performs region-growing segmentation by employing the hierarchical clustering procedure of CN-chain with the restriction that pixels in a cluster must be spatially contiguous. The "global" segmentor of the second stage, which has not spatial constraints for merging, clusters the segments resulting from the previous stage, using a context-free similarity measure. This study applied the multistage hierarchical clustering method to the data generated by band reduction, band selection and data compression. The classification results were compared with them using full bands.

Classification of land cover based on hyperspectral data is very challenging because typically tens of classes with uneven priors are involved, the inputs are high dimensional, and there is often scarcity of labeled data. Several researchers have observed that it is often preferable to decompose a multiclass problem into multiple two-class problems, solve each such subproblem using a suitable binary classifier, and then combine the outputs of this collection of classifiers in a suitable manner to obtain the answer to the original multiclass problem. This approach is taken by the popular error correcting output codes (ECOC) technique, as well by the binary hierarchical classifier (BHC). Classical techniques for dealing with small sample sizes include regularization of covariance matrices and feature reduction. In this paper we address the twin problems of small sample sizes and multiclass settings by proposing a feature reduction scheme that adaptively adjusts to the amount of labeled data available. This scheme can be used in conjunction with ECOC and the BHC, as well as other approaches such as round-robin classification that decompose a multiclass problem into a number of two (meta)-class problems. In particular, we develop the best-basis binary hierarchical classifier (BB-BHC) and best basis ECOC (BB-ECOC) families of models that are adapted to "small sample size" situations. Currently, there are few studies that compare the efficacy of different approaches to multiclass problems in general settings as well as in the specific context of small sample sizes. Our experiments on two sets of remote sensing data show that both BB-BHC and BB-ECOC methods are superior to their nonadaptive versions when faced with limited data, with the BB-BHC showing a slight edge in terms of classification accuracy as well as interpretability.

Multitemporal approaches using sequential data acquired over multiple years are essential for satisfactory discrimination between many land cover classes whose signatures exhibit seasonal trends. At any particular time, the response of several classes may be indistinguishable. Using the estimates of periodogram which are obtained from sequential images through FFT, multiple periodicities of the process have been incorporates into multitemporal classification. The Normalized Difference Vegetation Index (NDVI) was computed for five-day composites of the Advanced Very High Resolution Radiometer (AVHRR) imagery over Texas for 1995-2002 using a dynamic technique.

High dimensional inputs coupled with scarcity of labeled data are among the greatest challenges for classification of hyperspectral data. These problems are exacerbated if the number of classes is large. High dimensional output classes can often be handled effectively by decomposition into multiple two-(meta)class problems, where each sub-problem is solved using a suitable binary classifier, and outputs of this collection of classifiers are combined in a suitable manner to obtain the answer to the original multi-class problem. This approach is taken by the binary hierarchical classifier (BHC). The advantages of the BHC for output decomposition can be further exploited for hyperspectral data analysis by integrating a feature selection methodology with the classifier. Building upon the previously developed best bases BHC algorithm with greedy feature selection, a new method is developed that selects a subset of band groups within metaclasses using reactive tabu search. Experimental results obtained from analysis of Hyperion data acquired over the Okavango Delta in Botswana are superior to those of the greedy feature selection approach and more robust than either the original BHC or the BHC with greedy feature selection.

Classification of hyperspectral data is challenging because of high dimensionality inputs coupled with possible high dimensional outputs and scarcity of labeled information. Previously, a multiclassifier system was formulated in a binary hierarchical framework to group classes for accurate, rapid discrimination. In order to improve performance for small sample sizes, a new approach was developed that utilizes a feature reduction scheme which adaptively adjusts to the amount of labeled data available, while exploiting the fact that certain adjacent hyperspectral bands are highly correlated. The resulting best-basis binary hierarchical classifier (BB-BHC) family is thus able to address the "small sample size" problem, as evidenced by experimental results obtained from analysis of AVIRIS and Hyperion data acquired over Kennedy Space Center.

Located in northwestern Botswana and fed by the Okavango River originating in Angola's western highlands, the Okavango Delta is the world's largest inland delta. The floodwaters require approximately nine months to flow from the source to the bottom of the Delta due to the extremely low topographic relief. During the peak of flooding, the delta's area can expand to over 16,000 square kilometers, almost doubling the 9,000 square kilometers during the dry season. Investigation of the hydrologic cycle of the Okavango is relevant to studies ranging from fundamental research in climate change to exploration for groundwater to alleviate water shortages faced by local villages. Because the extent and inaccessibility of many areas of the Delta, the application of remote sensing technology is extremely attractive, if it can be shown to effectively map important landcover and geomorphological characteristics (or their surrogates). While results of previous investigations of the capability of multispectral sensors for characterizing drying gradients were promising, the increased dynamic range and additional bands of ALI data potentially provide greater capability for mapping flooding events and characterizing the ecological changes that are rapidly occurring in the Okavango Delta due to climate change and anthropogenic impacts. Further, Hyperion hyperspectral data should provide even greater discrimination of complex vegetation assemblages and improved characterization of seasonal changes in spectral response of vegetation to the annual flooding event. A series of eight near cloud-free EO-1 ALI and Hyperion acquisitions were obtained over Chief's Island (located in the center of the Delta) spanning the 2001 dry and flooding seasons.

INSAR data from the ERS-1/2 platforms are combined with multiple sets of data acquired by the NASA/JPL TOPSAR platform to obtain statistically optimal high-resolution estimates of topography over the Finke River Gorge in central Australia. The INSAR data are fused using a multiscale Kalman smoother. The estimated topography preserves the spatial resolution of the TOPSAR data while smoothing noise and providing estimates where there was no TOPSAR coverage. It is shown that the estimation error associated with the multiscale Kalman smoother is smaller than that obtained with a deterministic least squares approach.

A new, computationally efficient classification methodology was developed and implemented to classify Light Detection and Ranging (LIDAR) data as ground, vegetation, and man-made features (Weed 2001). The new procedure consists of several components that create ground, vegetation, and building surfaces, which are then used to classify the first and last reflection of each laser pulse. Ground and non-ground data are classified by adapting the concept of a lower envelope follower used to recover information in an amplitude modulated (AM) signal to the problem of extracting the ground surface from the LIDAR signal. The detected ground points include bare surface pixels and locations where the laser was able to penetrate the vegetation canopy, but exclude buildings and vegetation. Buildings are then classified by detecting the extended low gradient regions on their roofs. The first return LIDAR data points are used to accurately detect building edges distorted by multi-path errors in the last return LIDAR data. The combined roof and edge surfaces are then employed to threshold the first and last return LIDAR height values and detect the LIDAR points reflecting from buildings. Once the building points are classified, the vegetation points are extracted from the remaining LIDAR points using a mask of the regions where there were significant differences in the first and last return of the laser pulse. The technique is robust for classifying LIDAR data acquired over a range of terrains with different vegetation cover and types and sizes of buildings. It requires minimal user intervention for parameter selection.

Hurricane Iris, a Category Four hurricane on the Saffir-Simpson hurricane scale with winds exceeding 200 kph, made landfall in southern Belize, Central America, on October 8, 2001. Extensive wind damage occurred, including toppled and defoliated trees, and major losses to the local banana industry. Among the regions impacted by the storm was the Monkey River area located approximately 130 km south of Belize City. Imagery acquired on December 4, 2001, from NASA's Earth Observing-1 (EO-1) Advanced Land Imager (ALI) was used to analyze impacts on land cover/land use with emphasis on broadleaf forests. Comparisons were made with pre-hurricane Landsat TM data, in which 14 land cover/land use classes, including 6 classes of forests and savannah, 5 classes of wetlands and coastal lands, and 3 classes of developed land were classified.

This study has presented an approach to unsupervisedly estimate the number of classes and the parameters of defining the classes in order to train the classifier. Region growing segmentation and local fuzzy classification have been employed to find the sample classes that well represent the true image. The segmentation algorithm makes use of spatial contextual information in a hierarchical clustering procedure and multi-window operation using a pyramid-like structure to increase the computational efficiency. The fuzzy classification, which conducts classification by iteratively identifying expected maximum likelihood parameters of the class, is applied for the segmented regions in order to determine the sample classes. The maximum likelihood classifier has been used the unlabelled regions to assign them into one of a finite number of classes. The algorithm has been evaluated with simulated image data with various class patterns.

Due to advances in sensor technology, it is now possible to acquire hyperspectral data simultaneously in hundreds of bands. Algorithms that both reduce the dimensionality of the data sets and handle highly correlated bands are required to exploit the information in these data sets effectively. the authors propose a set of best-bases feature extraction algorithms that are simple, fast, and highly effective for classification of hyperspectral data. These techniques intelligently combine subsets of adjacent bands into a smaller number of features. Both top-down and bottom-up algorithms are proposed. The top-down algorithm recursively partitions the bands into two (not necessarily equal) sets of bands and then replaces each final set of bands by its mean value. The bottom-up algorithm builds an agglomerative tree by merging highly correlated adjacent bands and projecting them onto their Fisher direction, yielding high discrimination among classes. Both these algorithms are used in a pairwise classifier framework where the original C-class problem is divided into a set of (/sub 2//sup C/) two-class problems. The new algorithms (1) find variable length bases localized in wavelength, (2) favor grouping highly correlated adjacent bands that, when merged either by taking their mean or Fisher linear projection, yield maximum discrimination, and (3) seek orthogonal bases for each of the (/sub 2//sup C/) two-class problems into which a C-class problem can be decomposed. Experiments on an AVIRIS data set for a 12-class problem show significant improvements in classification accuracies while using a much smaller number of features.

Interferometric synthetic aperture radar (INSAR) and laser altimeter (LIDAR) systems are both widely used for mapping topography. INSAR can map extended areas but accuracies are limited over vegetated regions, primarily because the observations are not measurements of true surface topography. The measurements correspond to a height above the true surface that depends on both the sensor and the vegetation. Conversely, topography from LIDAR is very accurate, but coverage is limited to smaller regions. The authors demonstrate how these technologies can be used synergistically. First, the authors determine surface elevations and vegetation heights from dual-baseline INSAR data by inverting an INSAR scattering model. The authors then combine sparse LIDAR observations with the INSAR inversion results to improve the estimates of ground elevations and vegetation heights. This is accomplished via a multiresolution Kalman Filter that provides both the estimates and a measure of their uncertainty at each location. Combining data from the two sensors provides estimates that are more accurate than those obtained from INSAR alone yet have dense, extensive coverage, which is difficult to obtain with LIDAR. Contributions of this work include (1) combining physical modeling with multiscale estimation to accommodate nonlinear measurement-state relationships and (2) improving estimates of ground elevations and vegetation heights for remote sensing applications.

Conventional methods for feature selection use some kind of separability criteria or classification accuracy for computing the relevance of a feature subset to the classification task. In two-class problems, this approach may be suitable, but for problems such as character recognition with 26 classes, these feature selection algorithms are often faced with complex tradeoffs among efficacy of features for separating different subsets of classes. We propose a class-pair based feature selection algorithm which, in conjunction with mixture modeling technique, provides significantly superior results for differentiating a large number of classes, even when the class priors vary considerably. This technique is applied to multisensor NASA/JPL remote sensing AIRSAR data for characterizing 11 types of land cover. The proposed polychotomous approach not only gives improved test accuracy, but also reduces the number of features used. Important domain information can be derived from the features selected for different class pairs and the distance measure between these class pairs.

AIRSAR and TOPSAR data were acquired over the wetlands of Bolivar Peninsula along the Gulf coast of Texas for mapping land cover types and topographic features such as beach ridges, dunes, and relict storm features. Classification of land cover over this wetlands and uplands environment is difficult because of the similarity of spectral signatures of the vegetation types. In addition, because the distribution of vegetation communities in coastal marshes is strongly related to salinity, which in turn is largely dictated by frequency and duration of inundation, surface topography is critical to determination of the vegetation characteristics at any location. The potential advantages of multisensor classification, including, in particular, topographic information from a TOPSAR DEM are investigated. An approach which employs a class dependent feature selection procedure in conjunction with pairwise Bayesian classifiers is proposed and applied to the polarimetric and interferometric SAR data.

We introduce the Anisotropic Diffusion Pyramid (ADP), a structure for multiresolution image processing. We also develop the ADP for use in region-based segmentation. The pyramid is constructed using the anisotropic diffusion equations, creating an efficient scale-space representation. Segmentation is accomplished using pyramid node linking. Since anisotropic diffusion preserves edge localization as the scale is increased, the region boundaries in the coarse-to-fine ADP segmentation are accurately delineated. An application to segmentation of remotely sensed data is provided. The results of ADP segmentation are compared to Gaussian-based pyramidal segmentation. The examples show that the ADP has a superior ability to subdivide the image into integral groupings, minimizing the error in boundary localization and in pixel intensity.>

A multistage algorithm which makes use of spatial contextual information in a hierarchical clustering procedure has been developed for unsupervised image segmentation. A Markov random field model is employed to enforce local spatial smoothness, while the maximum entropy principle is utilized to quantify global smoothness in the image processing. A multiwindow approach implemented in a pyramid-like data structure which uses a boundary blocking operation is employed to increase computational efficiency.>

An unsupervised region-based image segmentation algorithm implemented with a pyramid structure has been developed. Rather than depending on traditional local splitting and merging of regions with a similarity test of region statistics, the algorithm identifies the homogeneous and boundary regions at each level of the pyramid; the global parameters of each class are then estimated and updated with the values of the homogeneous regions represented at that level of the pyramid using mixture distribution estimation. The image is then classified through the pyramid structure. Classification results obtained for both simulated and SPOT imagery are presented.>