Sathishkumar Samiappan

Non-Uniform Random Feature Selection and Kernel Density Scoring With SVM Based Ensemble Classification for Hyperspectral Image Analysis

Traditional statistical classification approaches often fail to yield adequate results with Hyperspectral imagery (HSI) because of the high dimensional nature of the data, multimodal class distribution and limited ground truth samples for training. Over the last decade, Support Vector Machines (SVMs) and Multi-Classifier Systems (MCS) have become popular tools for HSI analysis. Random Feature Selection (RFS) for MCS is a popular approach to produce higher classification accuracies. In this study, we present a Non-Uniform Random Feature Selection (NU-RFS) within a MCS framework using SVM as the base classifier. We propose a method to fuse the output of individual classifiers using scores derived from kernel density estimation. This study demonstrates the improvement in classification accuracies by comparing the proposed approach to conventional analysis algorithms and by assessing the sensitivity of the proposed approach to the number of training samples. These results are compared with that of uniform RFS and regular SVM classifiers. We demonstrate the superiority of Non-Uniform based RFS system with respect to overall accuracy, user accuracies, producer accuracies and sensitivity to number of training samples.

Using unmanned aerial vehicles for high-resolution remote sensing to map invasive Phragmites australis in coastal wetlands

ABSTRACT The wetland plant species, Phragmites australis, is present on every continent except Antarctica. Both native and non-native subspecies thrive in the USA with the non-natives quickly displacing native wetland plants. Along the Gulf Coast, Phragmites grows in very dense stands, and at heights of greater than 4.6 m, is usually the tallest grass species in a wetland, estuary, and marsh ecosystems. Phragmites is known to alter the ecology of these wetland systems making them less suitable as habitat for many species of flora and fauna. Furthermore, Phragmites presents a navigation hazard to smaller boats by impairing visibility along shorelines and around bends of canals and rivers. Management efforts targeting non-native Phragmites rely heavily on accurately mapping invaded areas. Historically, mapping has been done through walking the perimeter of a stand with a Global Positioning System (GPS) unit, using satellite imagery, or through the use of aerial photography from manned aircraft. These methods are time consuming, are expensive, can have an inadequate resolution, and in some cases are prone to human error. In this work, an Unmanned Aerial System (UAS) was used to capture visible imagery to create a basin-wide distribution map of a large wetland along the US Pearl River delta in southeastern Louisiana. The imagery was collected in the summer and individual images were mosaicked to create a larger map. We then evaluated the use of texture analysis on the mosaics to automatically map the invasive. Specifically, Gabor filters, grey level co-occurrence matrices, segmentation-based fractal texture analysis, and wavelet-based texture analysis were compared for mapping the Phragmites. Our experimental results, conducted using the imagery we collected over four study areas (approximately 2250 ha) along the US Pearl River delta, indicate the proposed texture-based approach yields an average accuracy of 85%, an average kappa accuracy of 70%. These maps have shown to be very useful for resource managers to hasten the eradication efforts of Phragmites.

Automated hyperspectral imagery analysis via support vector machines based multi-classifier system with non-uniform random feature selection

Ground cover classification using remotely sensed hyperspectral data is a challenging pattern recognition problem. The small (and expensive to collect) training sample sizes exacerbate the curse-of-dimensionality problem that already exists with such high dimensional feature spaces. However, Support Vector Machine (SVM) classifiers have been demonstrated to be better at handling such situations compared to other statistical classifiers. Recently, multi-classifier systems and a uniform random feature selection have proved to be very effective for hyperspectral image classification. In this paper, a support vector machines based multi-classifier system with non-uniform (spectrally-constrained) random feature selection is presented. We propose two approaches to perform such a non-uniform random-feature selection. Experimental results with the AVIRIS Indian Pines hyperspectral data demonstrate that the proposed approach outperforms regular random feature selection based on a uniform distribution.

NASA's upcoming HyspIRI mission — precision vegetation mapping with limited ground truth

Imaging spectrometers can acquire spectrally resolved images over a wide range of the electromagnetic spectrum. Availability of this rich spectral information makes it possible to design classification systems that can perform very accurate ground cover classification and target recognition. The Hyperspectral Infrared Imager (HyspIRI) — a National Research Council (NRC) decadal survey mission is much anticipated by researchers to aid in answering a wide variety of global ecological and anthropological questions. In this work, we study the efficacy of HyspIRI observations in precision vegetation mapping applications under limited ground truth availability. In particular, sensitivity to mixed pixel conditions, temporal misalignments and the amount of training (ground-truth) data are studied for various state-of-the-art classification systems. This study will provide valuable insight on the relationship between the quality and quantity of available ground truth and performance of classification systems with these HyspIRI observations.

Fusion of synthetic aperture radar and hyperspectral imagery to detect impacts of oil spill in Gulf of Mexico

The Deepwater Horizon blowout in the Gulf of Mexico resulted in one of the largest accidental oil disasters in U.S. history. NASA acquired radar and hyperspectral imagery and made them available to the scientific community for analyzing impacts of the oil spill. In this study, we use the L-band quad-polarized radar data acquired by Unmanned Aerial Vehicle Synthetic Aperture Radar (UAVSAR) and Hyperspectral Imagery (HSI) from the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) optical sensor. The main objective of this research is to apply fusion techniques on polarimetric radar and hyperspectral imagery to investigate the benefit of fusion for improved classification of coastal vegetation contaminated by oil. In this approach, fusion is implemented at the pixel level by concatenating the hyperspectral data with the high resolution SAR data and analyze the fused data with Support Vector Machine (SVM) classification algorithm.

Semi-supervised co-training and active learning framework for hyperspectral image classification

Hyperspectral imaging enables detailed ground cover classification with hundreds of spectral bands at each pixel. Rich spectral information can be a drawback since supervised classification of a hyperspectral image requires a balance between the number of training samples and its dimension. Achieving this balance requires a large number of training or ground truth samples, which is generally difficult, expensive and time-consuming. This led researchers to explore the use of semi-supervised learning techniques where new training samples (unlabeled) are obtained from a small set of available labeled samples without significant effort. In this paper, we propose a semi-supervised approach which adapts active learning to a co-training framework in which the algorithm automatically selects new training samples from abundant unlabeled pixels. Efficacy of the proposed approach is validated using a probabilistic support vector machine classifier. Our experimental results with an Indian Pines hyperspectral image collected by the National Aeronautics and Space Administration Jet Propulsion Laboratorys Airborne Visible-Infrared Imaging Spectrometer indicate that the use of this co-training based approach represents promising strategy in the context of hyperspectral image classification.

Mapping of invasive phragmites common reed in Gulf of Mexico coastal wetlands using multispectral imagery and small unmanned aerial systems

ABSTRACT In coastal wetlands of the Gulf of Mexico, the invasive plant species Phragmites australis (common reed) rapidly alters the ecology of a site by shifting plant communities from heterogeneous mixtures of plant species to homogenous stands of Phragmites. Phragmites grows in very dense stands at an average height of 4.6 m and outcompetes native plants for resources. To restore affected wetlands, resource managers require an accurate map of Phragmites locations. Previous studies have used satellite and manned aircraft-based remote-sensing images to map Phragmites in relatively large areas at a coarse scale; however, low-altitude high-spatial-resolution pixel-based classification approaches would improve the mapping accuracy. This study explores the supervised classification methods to accurately map Phragmites in the coastal wetlands at the delta of the Pearl River in Louisiana and Mississippi, USA, using high-resolution (8 cm ground sample distance; GSD) multispectral imagery collected from a small unmanned aerial system platform at an altitude of 120 m. We create a map through pixel-based Support Vector Machines (SVM) classification using blue, green, red, red edge, and near-infrared spectral bands along with a digital surface model (DSM), vegetation indices, and morphological attribute profiles (MAPs) as features. This study also demonstrates the effects of different features and their usefulness in generating an accurate map of Phragmites locations. Accuracy assessment based on a) a subset of training/testing samples (to show classifier performance) and b) the entire ground reference (GR) map (to show the quality of mapping) is demonstrated. Kappa, overall accuracy (OA), class accuracies, and their confidence intervals (CIs) are reported. An OA of 91% and kappa of 63 is achieved. The results of this study indicate that features such as MAPs are very useful in accurately mapping invasive Phragmites compared with existing region-based approaches.

Branch and bound based feature elimination for support vector machine based classification of hyperspectral images

Feature selection (FS) is a classical combinatorial problem in pattern recognition and data mining. It finds major importance in classification and regression scenarios. In this paper, a hybrid approach that combines branch-and-bound (BB) search with Bhattacharya distance based feature selection is presented for classifying hyperspectral data using Support Vector Machine (SVM) classifiers. The performance of this hybrid approach is compared to another hybrid approach that uses genetic algorithm (GA) based feature selection in place of BB. It is also compared to baseline SVMs with no feature reduction. Experimental results using hyperspectral data show that under small sample size situations, BB approach performs better than GA and SVM with no feature selection.

Classifying common wetland plants using hyperspectral data to identify optimal spectral bands for species mapping using a small unmanned aerial systems — A case study

Plant species monitoring in wetland ecosystems is crucial for preservation of water quality and many other ecological functions. Difficulties associated with conducting field work (i.e., hazardous terrain) can affect the ability of resource managers to correctly identify wetland plant species in a timely fashion. Thus, in wetland sites where access can be difficult, differentiation of plant species from aerial imagery can be of use. In this work, we studied laboratory produced hyperspectral data to classify various wetland plants thereby determining optimal multispectral bands that can distinguish these species. We then used an unmanned aerial systems (UAS) mountable 5-band multispectral sensor to classify 11 different wetland plant species to determine effectiveness of the multispectral bands chosen by the preliminary hyperspectral analysis. Classification experiments on the hyperspectral data produced the overall accuracies of 94–97% with 20% training samples. Experiments on the UAS collected multispectral data produced overall accuracy of 75% when considering four classes and 58% overall accuracy when considering 11 classes.

Probabilistic Neural Network and Wavelet Transform for Mapping of Phragmites Australis Using Low Altitude Remote Sensing

Phragmites australis (common reed) commonly found in the coastal wetlands can rapidly alter the ecology of these systems by outcompeting native plant species for resources. Identifying and mapping Phragmites can help resource managers to restore affected wetlands. In this work, we use probabilistic neural network with wavelet texture features for mapping regions with Phragmites in visible spectrum imagery acquired at low altitude with small unmanned aerial system. Evaluation study was conducted with imagery acquired in the delta of the Pearl River located in southeastern Louisiana and southwestern Mississippi, United States of America. In comparison to state-of-the-art, our approach presented improvements in several statistical variables such as overall accuracy and kappa value. Furthermore, we show that the remaining omission and commission errors with this technique are generally located along boundaries of patches with Phragmites, which reduces unnecessary efforts for resource managers while searching for nonexistent patches.