AVIRIS Research Articles

Improved Classification Method Based on the Diverse Density and Sparse Representation Model for a Hyperspectral Image.

To solve the small sample size (SSS) problem in the classification of hyperspectral image, a novel classification method based on diverse density and sparse representation (NCM_DDSR) is proposed. In the proposed method, the dictionary atoms, which learned from the diverse density model, are used to solve the noise interference problems of spectral features, and an improved matching pursuit model is presented to obtain the sparse coefficients. Airborne hyperspectral data collected by the push-broom hyperspectral imager (PHI) and the airborne visible/infrared imaging spectrometer (AVIRIS) are applied to evaluate the performance of the proposed classification method. Results illuminate that the overall accuracies of the proposed model for classification of PHI and AVIRIS images are up to 91.59% and 92.83% respectively. In addition, the kappa coefficients are up to 0.897 and 0.91.

An analysis of atmospheric water vapor variations over a complex agricultural region using airborne imaging spectrometry.

Understanding atmospheric water vapor patterns can inform regional understanding of water use, climate patterns and hydrologic processes. This research uses Airborne Visible Infrared Imaging Spectrometer (AVIRIS) reflectance and water vapor imagery to investigate spatial patterns of water vapor in California’s Central Valley on a June date in 2013, and 2015, and relates these patterns to surface characteristics and atmospheric properties. We analyze water vapor imagery at two scales: regional and agricultural field, to examine how the slope, intercept, and trajectory of water vapor interact with the landscape in a highly diverse and complex agricultural setting. At the field scale, we found significant quadratic relationships between water vapor slope and wind magnitude in both years (p<0.001). Results showed a positive correlation between crop water use and the frequency with which crops showed directional agreement between wind and water vapor (r = 0.23). At the regional scale, we found patterns of water vapor that indicate advection of moisture across the scene. Water vapor slope was inversely correlated to field size with correlations of -0.37, and -0.28 for 2013 and 2015. No correlation was found between green vegetation fraction and vapor slope (r = 0.001 in 2013, r = 0.02 in 2015), but a weak correlation was found for the intercept (r = 0.11 in 2013, r = 0.26 in 2015). These results lead us to conclude that accumulation of water vapor above fields in these scenes is observable with AVIRIS-derived water vapor imagery whereas advection at the field level was inconsistent. Based on these results, we identify new opportunities to use and apply water vapor imagery to advance our understanding of hydro-climatic patterns and applied agricultural water use.

Hyperspectral image classification using graph-based wavelet transform

ABSTRACTGraph-based methods are developed to efficiently extract data information. In particular, these methods are adopted for high-dimensional data classification by exploiting information residing on weighted graphs. In this paper, we propose a new hyperspectral texture classifier based on graph-based wavelet transform. This recent graph transform allows extracting textural features from a constructed weighted graph using sparse representative pixels of hyperspectral image. Different measurements of spectral similarity between representative pixels are tested to decorrelate close pixels and improve the classification precision. To achieve the hyperspectral texture classification, Support Vector Machine is applied on spectral graph wavelet coefficients. Experimental results obtained by applying the proposed approach on Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) and Reflective Optics System Imaging Spectrometer (ROSIS) datasets provide good accuracy which could exceed 98.7%. Compared to other famous classification methods as conventional deep learning-based methods, the proposed method achieves better classification performance. Results have shown the effectiveness of the method in terms of robustness and accuracy.

Portability Study of an OpenCL Algorithm for Automatic Target Detection in Hyperspectral Images

In the last decades, the problem of target detection has received considerable attention in remote sensing applications. When this problem is tackled using hyperspectral images with hundreds of bands, the use of high-performance computing (HPC) is essential. One of the most popular algorithms in the hyperspectral image analysis community for this purpose is the automatic target detection and classification algorithm (ATDCA). Previous research has already investigated the mapping of ATDCA on HPC platforms such as multicore processors, graphics processing units (GPUs), and field-programmable gate arrays (FPGAs), showing impressive speedup factors (after careful fine-tuning) that allow for its exploitation in time-critical scenarios. However, the lack of standardization resulted in most implementations being too specific to a given architecture, eliminating (or at least making extremely difficult) code reusability across different platforms. In order to address this issue, we present a portability study of an implementation of ATDCA developed using the open computing language (OpenCL). We focus on cross-platform parameters such as performance, energy consumption, and code design complexity, as compared to previously developed (hand-tuned) implementations. Our portability study analyzes different strategies to expose data parallelism as well as enable the efficient exploitation of complex memory hierarchies in heterogeneous devices. We also conduct an assessment of energy consumption and discuss metrics to analyze the quality of our code. The conducted experiments—using synthetic and real hyperspectral data sets collected by the Hyperspectral Digital Imagery Collection Experiment (HYDICE) and NASA’s Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS)—demonstrate, for the first time in the literature, that portability across different HPC platforms can be achieved for real-time target detection in hyperspectral missions.

Satellite data cloud detection using deep learning supported by hyperspectral data

ABSTRACTDeep learning methods can play an important role in satellite data cloud detection. The number and quality of training samples directly affect the accuracy of cloud detection based on deep learning. Therefore, selecting a large number of representative and high-quality training samples is a key step in cloud detection based on deep learning. For different satellite data sources, choosing sufficient and high-quality training samples has become an important factor limiting the application of deep learning in cloud detection. This paper presents a fast method for obtaining high-quality learning samples, which can be used for cloud detection of different satellite data with deep learning methods. AVIRIS (Airborne Visible Infrared Imaging Spectrometer) data, which have 224 continuous bands in the spectral range from 400–2500 nm, are used to provide cloud detection samples for different types of satellite data. Through visual interpretation, a sufficient number of cloud and clear sky pixels are selected from the AVIRIS data to construct a hyperspectral data sample library, which is used to simulate different satellite data (such as data from the Moderate Resolution Imaging Spectroradiometer (MODIS) and Landsat Operational Land Imager (OLI) satellites) as training samples. This approach avoids selecting training samples for different satellite sensors. Based on the Keras deep learning framework platform, a backpropagation (BP) neural network is employed for cloud detection from Landsat 8 OLI, National Polar-orbiting Partnership (NPP) Visible Infrared Imaging Radiometer Suite (VIIRS) and Terra MODIS data. The results are compared with cloud coverage results interpreted via artificial vision. The results demonstrate that the algorithm achieves good cloud detection results for the above data, and the overall accuracy is greater than 90%.

Spectral and Radiometric Calibration of the Next Generation Airborne Visible Infrared Spectrometer (AVIRIS-NG)

We describe advanced spectral and radiometric calibration techniques developed for NASA’s Next Generation Airborne Visible Infrared Imaging Spectrometer (AVIRIS-NG). By employing both statistically rigorous analysis and utilizing in situ data to inform calibration procedures and parameter estimation, we can dramatically reduce undesirable artifacts and minimize uncertainties of calibration parameters notoriously difficult to characterize in the laboratory. We describe a novel approach for destriping imaging spectrometer data through minimizing a Markov Random Field model. We then detail statistical methodology for bad pixel correction of the instrument, followed by the laboratory and field protocols involved in the corrections and evaluate their effectiveness on historical data. Finally, we review the geometric processing procedure used in production of the radiometrically calibrated image data.

NON-LINEAR AUTOENCODER BASED ALGORITHM FOR DIMENSIONALITY REDUCTION OF AIRBORNE HYPERSPECTRAL DATA

Abstract. Hyperspectral remote sensing is an advanced remote sensing technology that enhances the ability of accurate classification due to presence of narrow contiguous bands. The large number of continuous bands present in hyperspectral data introduces the problem of computational complexity due to presence of redundant information. There is a need for dimensionality reduction to enhance the ability of users for better characterization of features. Due to presence of high spectral correlation in the hyperspectral datasets, optimum de-correlation technique is required which transforms the hyperspectral data to lower dimensions without compromising with the desirable information present in the data. In this paper, focus has been to reduce the spectral dimensionality problem. So, this research aimed to develop computationally efficient non-linear autoencoder algorithm taking the advantage of non-linear properties of hyperspectral data. The proposed algorithm was applied on airborne hyperspectral image of Airborne Visible Infrared Imaging Spectrometer - Next Generation (AVIRIS-NG) over Anand region of Gujarat and the performance of the algorithm was evaluated. The Signal-to-Noise Ratio (SNR) increased from 22.78 dB to 48.48 dB with increase in number of nodes in bottleneck layer for reconstruction of image. Spectral distortion was also measured using Spectral Angle Mapper Algorithm (SAM), which reduced from 0.38 to 0.05 with increase in number of nodes in bottleneck layer up to 10. So, this algorithm was able to give good reconstruction of original image from the nodes present in the bottleneck layer.

Classifying California plant species temporally using airborne hyperspectral imagery

Accurate knowledge of seasonal and inter-annual distributions of plant species is required for many research and management agendas that track ecosystem health. Airborne imaging spectroscopy data have been used successfully to map plant species, but often only in a single season or over a limited spatial extent due to data availability. NASA's Hyperspectral Infrared Imager (HyspIRI) preparatory airborne campaign flew an imaging spectrometer from 2013 to 2015. This dataset captured a severe drought and thus provided the opportunity to evaluate species discrimination over an extreme range in environmental conditions. Here we evaluate the portability of image-based training data and accuracy of species discrimination. The imagery was acquired in the spring, summer, and fall seasons of 2013–2015 with the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Reference spectral libraries were developed with three sets of spectra: spectra from a single image date, combining spectra from multiple dates (by season, by year, and from all dates), and creating leave-one-out (LOO) libraries that pooled spectra from all dates but one. Canonical discriminant analysis (CDA) was applied to reduce data dimensionality, and classification was performed using linear discriminant analysis (LDA). When only spectra from the same image date were used, plant species were classified with a mean kappa accuracy ranging between 0.80 and 0.86 for the nine dates. Seasonal and annual spectral libraries had comparable accuracies with mean kappa 0.79–0.83 and 0.78–0.83, respectively. Seasonal libraries performed slightly better than annual libraries for species because they better-incorporated changes in spectra due to phenology. Spectral libraries were not transferable across dates, with mean kappa accuracies dropping to 0.31–0.73 for LOO spectral libraries. These results emphasize that spectral libraries built from previously collected imagery may not be able to map plant species over new images accurately. Specifically, our results highlight the need to use reference spectra that adequately represent the phenological and biophysical status of the plant species within an image for accurate mapping. Our research provides relevant insight for advanced species-mapping techniques across broad spatial and temporal scales using imagery from sensors like HyspIRI.

A novel spatial recurrent neural network for hyperspectral imagery classification

Abstract. Hyperspectral images (HSIs) contain hundreds of spectral bands, providing high-resolution spectral information pertaining to the Earth’s surface. Additionally, abundant spatial contextual information can also be obtained simultaneously from a HSI. To characterize the properties of ground objects, classification is the most widely-used technology in the field of remote sensing, where each pixel in a HSI is assigned to a pre-defined class. Over the past decade, deep learning has attracted increasing attention in the machine-learning and computer-vision domains, due to its favourable performances for various types of tasks, and it has been successfully introduced to the remote-sensing community. Instead of utilizing the shallow features within in a given image, which is the approach that is generally adopted in other conventional classification methods, deep-learning algorithms can extract hierarchical features from raw HSI data. Within the deep-learning framework, recurrent neural networks (RNNs), which are able to encode sequential features, have exhibited promising capabilities and have achieved encouraging performances, especially for the natural-language processing and speech-recognition communities. As multi-temporal remote-sensing images can be readily obtained from increasing numbers of satellite and unmanned aircraft systems, and since analysis of such multi-temporal data comprises a critical issue within numerous research subfields, including land-cover and land-change analyses, and land-resource management, RNNs have been applied in recent studies in order to extract temporal sequential features from multi-temporal remote-sensing images for the purpose of image classification. Apart from using multi-temporal image datasets, RNNs can also be utilized on a single image, where the spectral feature/band of each individual pixel can be taken as a sequential feature for the input layer of RNNs. However, the application of such sequential feature extraction that relies on a single image still needs to be further investigated since applying RNNs to spectral bands will directly introduce more parameters that need to be optimized, consequently increasing the total training time.In this study, we propose a novel RNN-based HSI classification framework. In this framework, unlabelled pixels obtained from a single image are considered when constructing sequential features. Two spatial similarity measurements, referred to as pixel-matching and block-matching, respectively, are employed to extract pixels that are “similar” to the target pixel. Then, the sequential feature of the target pixel is constructed by exploiting several of the most “similar” pixels and ordering them based on their similarities to the target pixel. The aforementioned two schemes are advantageous, as unlabelled pixels within the given HSI are taken into consideration for similarity measurement and sequential feature construction for the RNN model. Moreover, the block-matching scheme also takes advantage of spatial contextual information, which has been widely utilized in spatial-spectral-based HSI classification methods. To evaluate the proposed methods, two benchmark HSIs are used, including a HSI collected over Pavia University, Italy by the airborne Reflective Optics System Imaging Spectrometer (ROSIS) sensor, and an image acquired over the Salinas Valley, California, USA via the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) sensor. Spatio-temporally coincident ground-reference data accompanies each of these respective HSIs. In addition, the proposed methods are compared with three state-of-the-art algorithms, including support vector machine (SVM), the 1-dimensional convolutional neural network (1DCNN), and the 1-dimensional RNN (1DRNN).Experimental results indicate that our proposed methods achieve markedly better classification performance compared with the baseline algorithms on both datasets. For example, for the Pavia University image, the block-matching based RNN achieves the highest overall classification accuracy, with 94.32% accuracy, which is 9.87% higher than the next most accurate algorithm of the aforementioned three baseline methods, which in this case is the 1DCNN, with 84.45% overall accuracy. More specifically, the block-matching method performs better than the pixel-matching method in terms of both quantitative and qualitative assessments. Based on visual assessment/interpretation of the classification maps, it is apparent that “salt-and-pepper” noise is markedly alleviated; with block-matching, smoother classified images are generated compared with pixel-matching-based methods and the three baseline algorithms. Such results demonstrate the effectiveness of utilizing spatial contextual information in the similarity measurement.

Optimal estimation of spectral surface reflectance in challenging atmospheres

Optimal Estimation (OE) methods can simultaneously estimate surface and atmospheric properties from remote Visible/Shortwave imaging spectroscopy. Simultaneous solutions can improve retrieval accuracy with principled uncertainty quantification for hypothesis testing. While OE has been validated under benign atmospheric conditions, future global missions will observe environments with high aerosol and water vapor loadings. This work addresses the gap with diverse scenes from NASA's Next Generation Airborne Visible Infrared Imaging Spectrometer (AVIRIS-NG) India campaign. We refine atmospheric models to represent variable aerosol optical depths and properties. We quantify retrieval accuracy and information content for both reflectance and aerosols over different surface types, comparing results to in situ and remote references. Additionally, we assess uncertainty of maximum a posteriori solutions using linearized estimates as well as sampling-based inversions that more completely characterize posterior uncertainties. Principled uncertainty quantification can combine multiple spacecraft data products while preventing local environmental biases in future global investigations.

A unified approach to estimate land and water reflectances with uncertainties for coastal imaging spectroscopy

Coastal ecosystem studies using remote visible/infrared spectroscopy typically invert an atmospheric model to estimate the water-leaving reflectance signal. This inversion is challenging due to the confounding effects of turbid backscatter, atmospheric aerosols, and sun glint. Simultaneous estimation of the surface and atmosphere can resolve the ambiguity enabling spectral reflectance maps with rigorous uncertainty quantification. We demonstrate a simultaneous retrieval method that adapts the Optimal Estimation (OE) formalism of Rodgers (2000) to the coastal domain. We compare two surface representations: a parametric bio-optical model based on Inherent Optical Properties (IOPs); and an expressive statistical model that estimates reflectance in every instrument channel. The latter is suited to both land and water reflectance, enabling a unified analysis of terrestrial and aquatic domains. We test these models with both vector and scalar Radiative Transfer Models (RTMs). We report field experiments by two airborne instruments: NASA's Portable Remote Imaging SpectroMeter (PRISM) in an overflight of Santa Monica, California; and NASA's Next Generation Airborne Visible Infrared Imaging Spectrometer (AVIRIS-NG) in an overflight of the Wax Lake Delta and lower Atchafalaya River, Louisiana. In both cases, in situ validation measurements match remote water-leaving reflectance estimates to high accuracy. Posterior error predictions demonstrate a closed account of uncertainty in these coastal observations.

Integrating multi-sensor remote sensing and species distribution modeling to map the spread of emerging forest disease and tree mortality

Forest ecosystems have been increasingly affected by a variety of disturbances, including emerging infectious diseases (EIDs), causing extensive tree mortality in the Western United States. Especially over the past decade, EID outbreaks occurred more frequently and severely in forest landscapes, which have killed large numbers of trees. While tree mortality is observable from remote sensing, its symptom may be associated with both disease and non-disease disturbances (e.g., wildfire and drought). Species distribution modeling is widely used to understand species spatial preferences for certain habitat conditions, which may constrain uncertain remote sensing approaches due to limited spatial and spectral resolution. In this study, we integrated multi-sensor remote sensing and species distribution modeling to map disease-caused tree mortality in a forested area of 80,000 ha from 2005 to 2016. We selected sudden oak death (caused by pathogen P. ramorum) as a case study of a rapidly spreading emerging infectious disease, which has killed millions of oak (Quercus spp.) and tanoak (Lithocarpus densiflorus) in California over the past decades. To balance the needs for fine-scale monitoring of disease distribution patterns and satisfactory coverage at broad scales, our method applied spectral unmixing to extract sub-pixel disease presence using yearly Landsat time series. The results were improved by employing the probability of disease infection generated from a species distribution model. We calibrated and validated the method with image samples from high-spatial resolution NAIP (National Agriculture Imagery Program), and hyperspectral AVIRIS (Airborne Visible/Infrared Imaging Spectrometer) sensors, Google Earth® imagery, and field observations. The findings reveal an annual sudden oak death infection rate of 7% from 2005 to 2016, with overall mapping accuracies ranging from 76% to 83%. The integration of multi-sensor remote sensing and species distribution modeling considerably reduced the overestimation of disease effects as compared to the use of remote sensing alone, leading to an average of 26% decrease in detecting disease-affected trees. Such integration strategy proved the effectiveness of mapping long-term, disease-caused tree mortality in forest landscapes that have experienced multiple disturbances.

Unsupervised hyperspectral band selection using ranking based on a denoising error matching approach

ABSTRACT In this paper, we focus on utilizing the image denoising method for ranking of significant bands in hyperspectral imagery. We make use of the fact that the denoising error of bands varies with the significant information content of the bands in hyperspectral imagery. The denoising error is computed for each band individually and compared using a matching parameter with the denoising error of a reference image. The reference image is selected to be the first principal component corresponding to the maximum information. Three matching parameters including mutual information (MI), correlation coefficient (r) and the structural similarity index (SSIM) were used for ranking the bands based on the match with the denoising error of the reference image. The proposed algorithm is tested using three datasets, namely, Indian Pines, Salinas and Dhundi. The Indian Pines and Salinas datasets were acquired from the Airborne Visible Infrared Imaging Spectrometer (AVIRIS) sensor and comprised rural and agricultural area. The Dhundi dataset of Hyperion comprises mostly of features corresponding to snow-covered mountainous regions. To assess the accuracy of the proposed method, a supervised classification was carried out using a random forest classifier with 20% training pixels selected randomly from the ground reference. The proposed method yielded significantly better results determined by the kappa coefficient (κ) of 0.756, 0.910 and 0.996 for the Indian Pines, Salinas and Dhundi datasets, respectively, over several other state of the art methods. The classification results of the proposed method also yielded better results than those obtained by the state-of-the-art methods for hyperspectral band selection.

Imaging spectrometry-derived estimates of regional ecosystem composition for the Sierra Nevada, California

The composition of the plant canopy is a key attribute of terrestrial ecosystems, influencing the fluxes of carbon, water, and energy between the land surface and the atmosphere. Terrestrial ecosystem and biosphere models, which are used to predict how ecosystems are expected to respond to changes in climate, atmospheric CO2, and land-use change, require accurate representations of plant canopy composition at large spatial scales. The ability to accurately specify plant canopy composition is important because it determines the physiological and ecological properties of plants (such as leaf photosynthetic capacity, patterns of plant carbon allocation and tissue turnover, and the resulting dynamics of plant demography) that govern the biophysical and biogeochemical functioning of ecosystems. Traditionally, plant canopy composition has been represented in a coarse-grained manner within terrestrial biosphere models, with ecosystems being comprised of a single plant functional type (PFT). However, models are increasingly seeking to represent fine-scale spatial variation in plant functional diversity. In this study, we show how imaging spectrometry measurements can provide spatially-comprehensive estimates of within-biome heterogeneity in PFT composition across a functionally diverse and topographically heterogeneous ~710 km2 area in the Southern Sierra Mountains of California. AVIRIS (Airborne Visible Infrared Imaging Spectrometer) data at 18 m resolution from the recent HyspIRI Preparatory Mission (Hyperspectral InfraRed Imager) were used to estimate the sub-pixel fractions of seven PFTs represented in the ED2 terrestrial biosphere model: Shrub, Oak, Western Hardwood, Western Pine, Cedar/Fir, and High-elevation Pine, plus a Grass/NPV (Non-Photosynthetic Vegetation) fraction using Multiple Endmember Spectral Mixture Analysis (MESMA). ED2 is an individual-based terrestrial biosphere model capable of representing fine-scale sub-pixel ecosystem heterogeneity. Our results show that this methodology captures important elevation-related shifts in canopy composition that occur within the study area that are not resolved by existing multi-spectral land-cover products. These estimates modestly improved when the putative PFT endmembers considered in the mixture analysis were constrained using available geospatial data about the presence and absence of the PFTs in particular areas: the average RMSEs (root-mean-square errors) with the geospatially-constrained versus conventional method were 11.3% and 11.9% respectively, with larger reductions in the bias (i.e. mean error) in the abundances of Oak, Cedar/Fir, and Western Hardwood PFTs (ranging from 2.0% to 7.8%). At the hectare scale around four flux towers in the Southern Sierra Mountains, the overall composition improved from an RMSE of 18.2% (5.0–24.2% for individual PFTs) to an RMSE of 9.5% (3.3–13.2% for individual PFTs). Downgrading AVIRIS to 30 m resolution resulted in a reduction in accuracy of the constrained method to an RMSE of 12.7% (0–23.7%) with <1% change in the bias for all tree and shrub PFTs. Our results demonstrate that imaging spectrometry measurements from planned satellite missions such as HyspIRI, EnMAP (Environmental Mapping and Analysis Program), and HISUI (Hyper-spectral Imager SUIte) can provide important and much-needed information about fine-scale heterogeneity in the composition of plant canopies for constraining and improving terrestrial ecosystem and biosphere model simulations of regional- and global-scale vegetation dynamics and function.

Forest health estimation in Sholayar Reserve Forest, Kerala using AVIRIS-NG hyperspectral data

The present study deals with analyzing forest health, its parameters, and suitability of hyperspectral data for vegetation health-related studies. Sholayar reserve forest in Kerala has a huge reserve of equatorial moist evergreen forest and demands preservation in every respect. Due to increased human interferences coupled with possible climate change, its health is undergoing a stage of deterioration. Stress levels in the canopy were assessed using a number of stress-related pigments. Detailed study of vegetation response to canopy leaf pigments have been carried out in the study. Airborne Visible Infrared Imaging Spectrometer Next Generation (AVIRIS-NG) data provides immense possibilities to study a number of stress-related pigments like anthocyanin, carotenoid, lignin, chlorophyll-a, b etc. Dominant species in these forests are Holigarna arnottiana, Grevillea robusta, Grewia tiliifolia, Syzygium cumini, Alstonia Scholaris, Cinnamomum verum, Artocarpus heterophyllus, Bischofia javanica, Mangifera indica, Bombax ceiba, Anogeissus latifolia, Terminalia paniculata etc. Apart from luscious natural vegetation, plantation of teak (Tectona Grandis), rubber (Hevea brasiliensis), tea (Camellia sinensis), Coffee (Coffee Arabica), Palm-Oil tree (Elaeis guineensis) etc. also exists. Field data pertaining to one of the selected pigments was correlated with remotely sensed pigment estimates. Correlation of field measured chlorophyll concentration and EVI showed R2 = 0.421. Similarly, the anthocyanin index showed a correlation of R2 = 0.319. In the Sholayar Reserve Forest (493.0 km2) an area of 141.0 km2 was found to be in a healthy state. Whereas about 218.0 km2 of area exhibit moderately healthy condition and 77.0 km2 area was in the least healthy state.

A Novel Tri-Training Technique for the Semi-Supervised Classification of Hyperspectral Images Based on Regularized Local Discriminant Embedding Feature Extraction

This paper introduces a novel semi-supervised tri-training classification algorithm based on regularized local discriminant embedding (RLDE) for hyperspectral imagery. In this algorithm, the RLDE method is used for optimal feature information extraction, to solve the problems of singular values and over-fitting, which are the main problems in the local discriminant embedding (LDE) and local Fisher discriminant analysis (LFDA) methods. An active learning method is then used to select the most useful and informative samples from the candidate set. In the experiments undertaken in this study, the three base classifiers were multinomial logistic regression (MLR), k-nearest neighbor (KNN), and random forest (RF). To confirm the effectiveness of the proposed RLDE method, experiments were conducted on two real hyperspectral datasets (Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) and Reflective Optics System Imaging Spectrometer (ROSIS)), and the proposed RLDE tri-training algorithm was compared with its counterparts of tri-training alone, LDE, and LFDA. The experiments confirmed that the proposed approach can effectively improve the classification accuracy for hyperspectral imagery.

Entropy-Mediated Decision Fusion for Remotely Sensed Image Classification

To better classify remotely sensed hyperspectral imagery, we study hyperspectral signatures from a different view, in which the discriminatory information is divided as reflectance features and absorption features, respectively. Based on this categorization, we put forward an information fusion approach, where the reflectance features and the absorption features are processed by different algorithms. Their outputs are considered as initial decisions, and then fused by a decision-level algorithm, where the entropy of the classification output is used to balance between the two decisions. The final decision is reached by modifying the decision of the reflectance features via the results of the absorption features. Simulations are carried out to assess the classification performance based on two AVIRIS (Airborne Visible/Infrared Imaging Spectrometer) hyperspectral datasets. The results show that the proposed method increases the classification accuracy against the state-of-the-art methods.

Using paired thermal and hyperspectral aerial imagery to quantify land surface temperature variability and assess crop stress within California orchards

Remote sensing can inform agricultural knowledge of crop water use through observation of land surface temperature, which can act as an indicator of plant function and health. This study uses remotely sensed data to quantify thermal variability within fruit and nut orchards during an intense drought period in California's Central Valley (2013–2015). First, fractions of green vegetation (GV), non-photosynthetic vegetation (NPV), and soil were derived for a variety of crop species using visible-shortwave infrared (VSWIR) spectra imaged by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Fractional estimates were then used to select thermal endmembers for each class using simultaneously collected MODIS/ASTER Airborne Simulator (MASTER) thermal imagery and a crop species map. Expected pixel temperatures of non-stressed crop fields were then modeled, and the per-pixel difference between measured and expected temperatures was calculated as a temperature residual. Crop residuals serve to capture variability in temperature that may be attributable to differences in crop health and/or management practices. We found multiple distinct thermal classes to exist across the study site. Furthermore, crop temperatures correlated to expected crop ET rates, and temperature residuals showed correlations to changes in crop yields during the study period. Further assessment of findings revealed an increase in temperature residuals during the study period that is consistent with increasing stress, likely linked to the progression of drought. The method presented here shows utility for regional agricultural analysis of crop water use and is particularly relevant for ECOSTRESS and the upcoming Surface Biology & Geology mission.

CLASS-BASED AFFINITY PROPAGATION FOR HYPERSPECTRAL IMAGE DIMENSIONALITY REDUCTION AND IMPROVEMENT OF MAXIMUM LIKELIHOOD CLASSIFICATION ACCURACY

Abstract This paper investigates an alternative classification method that integrates class-based affinity propagation (CAP) clustering algorithm and maximum likelihood classifier (MLC) with the purpose of overcome the MLC limitations in the classification of high dimensionality data, and thus improve its accuracy. The new classifier was named CAP-MLC, and comprises two approaches, spectral feature selection and image classification. CAP clustering algorithm was used to perform the image dimensionality reduction and feature selection while the MLC was employed for image classification. The performance of MLC in terms of classification accuracy and processing time is determined as a function of the selection rate achieved in the CAP clustering stage. The performance of CAP-MLC has been evaluated and validated using two hyperspectral scenes from the Airborne Visible Infrared Imaging Spectrometer (AVIRIS) and the Hyperspectral Digital Imagery Collection Experiment (HYDICE). Classification results show that CAP-MLC observed an enormous improvement in accuracy, reaching 94.15% and 96.47% respectively for AVIRIS and HYDICE if compared with MLC, which had 85.42% and 81.50%. These values obtained by CAP-MLC improved the MLC classification accuracy in 8.73% and 14.97% for these images. The results also show that CAP-MLC performed well, even for classes with limited training samples, surpassing the limitations of MLC.

Una herramienta para el análisis de índices espectrales para la detección remota de vegetación y cultivos utilizando imágenes hiperespectrales

Los requerimientos alimentarios en el mundo han aumentado, evidenciando la necesidad de mejorar las técnicas estándar de producción agrícola. Para abordar este problema, una alternativa de solución es la inclusión de elementos tecnológicos como el sensado remoto de vegetación y los cultivos a partir de imágenes hiperespectrales. El sensado remoto y las imágenes hiperespectrales son métodos no invasivos, que permiten monitorear grandes espacios de terreno en cantidades de tiempo reducidas. Estas características han hecho que el sensado remoto a partir de imágenes hiperespectrales sea una herramienta poderosa para desarrollo de procesos de agricultura de precisión. En este artículo se presenta una aplicación de software que permite generar y procesar índices espectrales de vegetación y sus respectivas imágenes de pseudo color, utilizando imágenes hiperespectrales. Las imágenes hiperespectrales utilizadas fueron tomadas de la base de datos del sensor Airborne Visible-Infrared Imaging Spectrometer (AVIRIS), diseñado por la NASA. El objetivo de la aplicación de software es mostrar diferentes elementos asociados con el monitoreo remoto de vegetación y cultivos a partir de imágenes hiperespectrales. Finalmente, se presentan pruebas funcionales para verificar el cumplimiento de los requisitos del software.