Jeffries-Matusita Distance Research Articles

A Comparative Analysis of Phenological Curves for Major Crops in Kansas

The goal of this research was to conduct an initial investigation into whether a time-series NDVI reference curve library for crops over a growing season for one year could be used to map crops for a different year. Time-series NDVI libraries of curves for 2001 and 2005 were investigated to ascertain whether or not the 2001 dataset could be used to map crops for 2005. The 2005 16-day composite MODIS 250 m NDVI data were used to extract NDVI values from 1,615 field sites representing alfalfa, corn, sorghum, soybeans, and winter wheat. A k-means cluster analysis of NDVI values from the field sites was performed to identify validation sites with time-series NDVI spectral profiles characteristic of the major crop types grown in Kansas. After completing the field site refinement process, there were 1,254 field sites retained for further analysis, referred to as "final" field sites. The methods employed to evaluate whether the MODIS-based NDVI profiles for major crops in Kansas are stable from year-to-year involved both graphical and statistical analyses. First, the time-series NDVI values for 2005 from the final field sites were aggregated by crop type and the crop NDVI profiles were then visually assessed and compared to the profiles of 2001 to ascertain if each crop's unique phenological pattern was consistent between the two years. Second, separability within each crop class in the time-series NDVI data between 2001 and 2005 was investigated numerically using the Jeffries-Matusita (JM) distance statistic. The results seem to suggest that time-series NDVI response curves for crops over a growing period for one year of valid ground reference data may be useful for mapping crops for a different year when minor temporal shifts in the NDVI values (resulting from inter-annual climate variations or changes in agricultural management practices) are taken into account.

Optimal Envisat advanced synthetic aperture radar image parameters for mapping and monitoring Sahelian floodplains

Floodplains in the Sahel region of Africa are of exceptional socio-economical and ecological importance. Due to their large extent and highly dynamic nature, monitoring these ecosystems can only be performed by means of remote sensing. The capability of the Envisat Advanced Synthetic Aperture Radar (ASAR) sensor to capture radar backscattering at various incident angles and with different polarization combinations, provides opportunities for improved wetland mapping and monitoring. However, little is known of the optimal image parameters, i.e. incident angle, polarization combination, and acquisition time. Backscatter σ° signatures of Land Use and Land Cover (LULC) classes in and around the Waza-Logone floodplain (Cameroon) were analyzed to determine these optimal image parameters. Based on Jeffries-Matusita (JM) distances between all LULC classes it was determined that best separation was obtained with images acquired in the middle of the flooding cycle at a steep incident angle. Furthermore, separability of cross-polarized images was higher than for co-polarized images. The combination of two and three ASAR Alternating Polarization images with highest separability were used as input for a LULC classification. Two methods were evaluated: Pixel-based Maximum Likelihood and object-based Nearest Neighbour (NN) classification. Best results were obtained with the object-based approach.

Spectral discrimination of papyrus vegetation ( Cyperus papyrus L.) in swamp wetlands using field spectrometry

Techniques for mapping and monitoring wetland species are critical for their sustainable management. Papyrus ( Cyperus papyrus L.) is one of the most important species-rich habitats that characterize the Greater St. Lucia Wetlands Park (GSWP) in South Africa. This paper investigates whether papyrus could be discriminated from its co-existing species using ASD field spectrometer data ranging from 300 nm to 2500 nm, yielding a total of 2151 bands. Canopy spectral measurements from papyrus and three other species were collected in situ in the Greater St. Lucia Wetlands Park, South Africa. A new hierarchical method based on three integrated analysis levels was proposed and implemented to spectrally discriminate papyrus from other species as well as to reduce and subsequently select optimal bands for the potential discrimination of papyrus. In the first level of the analysis using ANOVA, we found that there were statistically significant differences in spectral reflectance between papyrus and other species on 412 wavelengths located in different portions of the electromagnetic spectrum. Using the selected 412 bands, we further investigated the use of classification and regression trees (CART) in the second level of analysis to identify the most sensitive bands for spectral discrimination. This analysis yielded eight bands which are considered to be practical for upscaling to airborne or space borne sensors for mapping papyrus vegetation. The final sensitivity analysis level involved the application of Jeffries-Matusita (JM) distance to assess the relative importance of the selected eight bands in discriminating papyrus from other species. The results indicate that the best discrimination of papyrus from its co-existing species is possible with six bands located in the red-edge and near-infrared regions of the electromagnetic spectrum. Overall, the study concluded that spectral reflectance of papyrus and its co-existing species is statistically different, a promising result for the use of airborne and satellite sensors for mapping papyrus. The three-step hierarchical approach employed in this study could systematically reduce the dimensionality of bands to manageable levels, a move towards operational implementation with band specific sensors.

An Automatic System for the Analysis and Classification of Human Atrial Fibrillation Patterns from Intracardiac Electrograms

This paper presents an automatic system for the analysis and classification of atrial fibrillation (AF) patterns from bipolar intracardiac signals. The system is made up of: 1) a feature-extraction module that defines and extracts a set of measures potentially useful for characterizing AF types on the basis of their degree of organization; 2) a feature-selection module (based on the Jeffries-Matusita distance and a branch and bound search algorithm) identifying the best subset of features for discriminating different AF types; and 3) a support vector machine technique-based classification module that automatically discriminates the AF types according to the Wells' criteria. The automatic system was applied on 100 intracardiac AF signal strips and on a selection of 11 representative features, demonstrating: a) the possibility to properly identify the most significant features for the discrimination of AF types; b) higher accuracy (97.7% using the seven most informative features) than the traditional maximum likelihood classifier; and c) effectiveness in AF classification also with few training samples (accuracy = 88.3% with only five training signals). Finally, the system identifies a combination of indices characterizing changes of morphology of atrial activation waves and perturbation of the isoelectric line as the most effective in separating the AF types.

Using the Bootstrap Method for a Statistical Significance Test of Differences between Summary Histograms

Abstract A new method is proposed to compare statistical differences between summary histograms, which are the histograms summed over a large ensemble of individual histograms. It consists of choosing a distance statistic for measuring the difference between summary histograms and using a bootstrap procedure to calculate the statistical significance level. Bootstrapping is an approach to statistical inference that makes few assumptions about the underlying probability distribution that describes the data. Three distance statistics are compared in this study. They are the Euclidean distance, the Jeffries–Matusita distance, and the Kuiper distance. The data used in testing the bootstrap method are satellite measurements of cloud systems called “cloud objects.” Each cloud object is defined as a contiguous region/patch composed of individual footprints or fields of view. A histogram of measured values over footprints is generated for each parameter of each cloud object, and then summary histograms are accumulated over all individual histograms in a given cloud-object size category. The results of statistical hypothesis tests using all three distances as test statistics are generally similar, indicating the validity of the proposed method. The Euclidean distance is determined to be most suitable after comparing the statistical tests of several parameters with distinct probability distributions among three cloud-object size categories. Impacts on the statistical significance levels resulting from differences in the total lengths of satellite footprint data between two size categories are also discussed.

Caracterização de lesões intersticiais de pulmão em radiograma de tórax utilizando análise local de textura

OBJETIVO: Caracterizar lesões intersticiais em radiografias frontais de tórax, com base na análise de atributos estatísticos de textura, os quais permitem detectar sinais de anormalidades com natureza difusa. MATERIAIS E MÉTODOS: O esquema começa com a segmentação semi-automática dos campos pulmonares, sendo o contorno externo marcado manualmente, com posterior divisão automática de cada pulmão em seis regiões. O banco de imagens utilizado neste trabalho é composto por 482 regiões obtidas de exames contendo lesões e 324 regiões obtidas de exames normais. Os atributos de textura são extraídos automaticamente de cada uma dessas regiões e uma seleção das melhores combinações de atributos é feita através da distância Jeffries-Matusita. A classificação das regiões em normal ou suspeita é feita pela comparação com os k vizinhos mais próximos e o treinamento do classificador é baseado na técnica de treino e teste "half-half" e correlação cruzada. RESULTADOS: Os resultados obtidos foram analisados através do valor da área sob a curva ROC ("receiver operating characteristic"), a qual indica um sistema perfeito para uma área igual a 1. Os resultados forneceram uma área sob a curva ROC (A Z) igual a 0,887, com valores de sensibilidade igual a 0,804 e especificidade igual a 0,793. CONCLUSÃO: Os resultados indicam que o sistema de caracterização baseado em atributos de textura possui bom potencial para o auxílio ao diagnóstico de lesões intersticiais de pulmão.

Integrating genetic algorithm method with neural network for land use classification using SZ-3 CMODIS data*

Abstract This paper presents a methodology on land use mapping using CMODIS (Chinese Moderate Resolution Imaging Spectroradiometer) data on-board SZ-3 (Shenzhou 3) spacecraft. The integrated method is composed of genetic algorithm (GA) for feature extraction and neural network classifier for land use classification. In the data preprocessing, a moment matching method was adopted to remove the stripes in the images. Then by using the reproduction, crossover and mutation operators of GA based on the mechanism of “natural selection”, and with Jeffries-Matusita distance as its discriminate rule and the training samples, the optimal band combination for land use classification was obtained. To generate a land use map, the three layers back propagation neural network classifier is used for training the samples and classification. Compared with the Maximum Likelihood classification algorithm, the results show that the accuracy of land use classification is obviously improved by using our proposed method, the sel...

A simulation analysis of the detectability of understory burns in miombo woodlands

The miombo woodlands of southern Africa are one of the most extensively burned biomes in the tropics. The detectability of understory burns in these woodlands was assessed with a sensitivity analysis approach, based on a hybrid geometrical–optical radiative transfer model. Model input data were obtained from a variety of sources, including field biometry and spectroradiometry, and satellite data. The effects of variable tree percent cover, leaf area index, stand density, burn scar age, illumination and observation geometry, and spectral region, were taken into account. Detectability of understory burns was defined as the spectral separability of burned and unburned understory, measured with the Jeffries–Matusita distance, for all possible combinations of the green, red, and near-infrared channels of the Moderate Resolution Imaging Spectrometer (MODIS) sensor. Single channels, or pairwise combinations of channels perform poorly at detecting understory burns, but a large improvement in detectability is obtained for the combination of the three spectral domains. The detectability of understory burns is largely insensitive to the effects of stand structure and illumination/observation geometry, and depends primarily on burn scar age. Our results agree with those of previous satellite-based studies of burns scar detectability in African savanna woodlands.

Visual Method for Spectral Band Selection

We present a new method for performing band selection experiments with spectral data. This method allows for the visual inspection and assessment of the experiment results, and includes a statistical significance test. The method follows a standard feature selection approach in which a multivariate distance measure is used as a figure of merit in a search-optimization procedure. For this letter, we have chosen the Jeffries-Matusita distance between each sample and its immediate background. The band selection methodology uses either an exhaustive search over all possible combinations of 1-4 bands or sequential forward selection. To analyze the band selection results, we count the number of times that each band is selected as a member of the best set by the protocol, and we plot the results as a band frequency histogram. This allows us to visually discern spectral patterns that are not evident otherwise, and thus better assess the utility of each spectral band. We can compute band frequency histograms over individual classes of samples or over groups of classes. In addition, we can compute a significance statistic that gives us the probability that a given histogram is not the result of random band selection outcomes.

Evaluating the performance of multitemporal image compositing algorithms for burned area analysis

The main objective of this study was to compare the adequacy of various multitemporal image compositing algorithms to produce composite images suitable for burned area analysis. Satellite imagery from the NOAA Advanced Very High Resolution Radiometer (AVHRR) from three different regions (Portugal, central Africa, and South America) were used to compare six algorithms, two of which involve the sequential application of two criteria. Performance of the algorithms was assessed with the Jeffries-Matusita distance, to quantify spectral separability of the burned and unburned classes in the composite images. The ability of the algorithms to avoid the retention of cloud shadows was assessed visually with red-green-blue colour composites, and the level of radiometric speckle in the composite images was quantified with the Moran's I spatial autocorrelation statistic. The commonly used NDVI maximum value compositing procedure was found to be the least appropriate to produce composites to be used for burned area mapping, from all standpoints. The best spectral separability is provided by the minimum channel 2 (m2) compositing approach which has, however, the drawback of retaining cloud shadows. A two-criterion approach which complements m2 with maximization of brightness temperature in a subset of the data (m2M4) is considered the better method.

Optimization at multi-spectral land cover classification with fuzzy clustering and the Kohonen feature map

The number and structure of land cover classes separatable in a region on the basis of multi-spectral satellite images are usually known. The method presented here is capable of combining information from the spectral and nonspectral channels of multi-spectral images in a way that offers a possibility to cover all spectrally separable classes. The method also enables the determination of the representativeness of the available training areas with respect to these classes. In the course of the procedure, 5-35 classes are iteratively determined in succession for different channel combinations with the Kohonen feature map and the fuzzy c-means clustering algorithm. The maximum Jeffries-Matusita (JM) distance between these classes indicates the optimum class number of unambiguously separable classes for the entire multi-spectral image. In a simple step, the available training areas are grouped by the fuzzy c-means clustering algorithm. In this case, the maximum JM distance indicates the class number that ensures optimum separation of training areas. If this class number is smaller than the optimum class number determined for the entire scene, then further training areas must be defined in order to improve the results of a subsequent supervised land cover classification. The method's efficiency is demonstrated by the example of land cover classification for the region around Perpignan.

Classification and Change Detection Using Landsat TM Data: When and How to Correct Atmospheric Effects?

The electromagnetic radiation (EMR) signals collected by satellites in the solar spectrum are modified by scattering and absorption by gases and aerosols while traveling through the atmosphere from the Earth's surface to the sensor. When and how to correct the atmospheric effects depend on the remote sensing and atmospheric data available, the information desired, and the analytical methods used to extract the information. In many applications involving classification and change detection, atmospheric correction is unnecessary as long as the training data and the data to be classified are in the same relative scale. In other circumstances, corrections are mandatory to put multitemporal data on the same radiometric scale in order to monitor terrestrial surfaces over time. A multitemporal dataset consisting of seven Landsat 5 Thematic Mapper (TM) images from 1988 to 1996 of the Pearl River Delta, Guangdong Province, China was used to compare seven absolute and one relative atmospheric correction algorithms with uncorrected raw data. Based on classification and change detection results, all corrections improved the data analysis. The best overall results are achieved using a new method which adds the effect of Rayleigh scattering to conventional dark object subtraction. Though this method may not lead to accurate surface reflectance, it best minimizes the difference in reflectances within a land cover class through time as measured with the Jeffries–Matusita distance. Contrary to expectations, the more complicated algorithms do not necessarily lead to improved performance of classification and change detection. Simple dark object subtraction, with or without the Rayleigh atmosphere correction, or relative atmospheric correction are recommended for classification and change detection applications.

Optimization at multi-spectral land cover classification with fuzzy clustering and the Kohonen feature map

The number and structure of land cover classes separatable in a region on the basis of multi-spectral satellite images are usually known. The method presented here is capable of combining information from the spectral and nonspectral channels of multi-spectral images in a way that offers a possibility to cover all spectrally separable classes. The method also enables the determination of the representativeness of the available training areas with respect to these classes. In the course of the procedure, 5-35 classes are iteratively determined in succession for different channel combinations with the Kohonen feature map and the fuzzy c-means clustering algorithm. The maximum Jeffries-Matusita (JM) distance between these classes indicates the optimum class number of unambiguously separable classes for the entire multi-spectral image. In a simple step, the available training areas are grouped by the fuzzy c-means clustering algorithm. In this case, the maximum JM distance indicates the class number that ensures optimum separation of training areas. If this class number is smaller than the optimum class number determined for the entire scene, then further training areas must be defined in order to improve the results of a subsequent supervised land cover classification. The method's efficiency is demonstrated by the example of land cover classification for the region around Perpignan.

Comparison of Algorithms for Classifying Swedish Landcover Using Landsat TM and ERS-1 SAR Data

Sixteen landcover classes in a representative Swedish environment were analyzed and classified using one Landsat TM scene and seven ERS-1 SAR.PRI images acquired during 1993. Spectral and backscattering signature separabilities are analyzed using the Jeffries–Matusita distance measure to determine which combinations of channels/images contained the most information. Maximum likelihood, sequential maximum a posteriori (SMAP, a Bayesian image segmentation algorithm), and back propagation neural network classification algorithms were applied and their performances evaluated. Results of the separability analyses indicated that the multitemporal SAR data contained more separable landcover information than did the multispectral TM data; the highest separabilities were achieved when the TM and SAR data were combined. Classification accuracy evaluation results indicate that the SMAP algorithm outperformed the maximum likelihood algorithm which, in turn, outperformed the neural network algorithm. The best KAPPA values, using combined data, were 0.495 for SMAP, 0.445 for maximum likelihood, and 0.432 for neural network. Corresponding overall accuracy values were 57.1%, 52.4%, and 51.2%, respectively. A comparison between lumped crop area statistics with areal sums calculated from the classified satellite data gave the highest correspondence where the SMAP algorithm was used, followed by the maximum likelihood and neural network algorithms. Based on our application, we can therefore confirm the value of a multisource optical/SAR approach for analyzing landcover and the improvements to classification achieved using the SMAP algorithm.

Study of crop discrimination by simulated IRS 1C LISS-III data

A study on crop discrimination was carried out using simulated IRS 1C LISS-III data produced using visible (to simulate B2, B3) and NIR (to simulate B4) channels from SPOT and middle infrared (MIR) channel (to simulate B5) from TM over a previously investigated test site, characterized by multiple crops and small fields, in Sabarkantha district (Gujarat). The separability amongst dominant kharif season crops, namely, cotton, groundnut, maize, pigeonpea, between crops and various natural vegetation classes was investigated using Jeffries-Matusita (JM) distance, a pair-wise inter-class separability measure. The study highlighted the capability of simulated LISS-III data to be useful in identifying and labelling small fields and the 4-band data set (B2345, i.e., simulated LISS-III) to significantly improve the separability amongst various crops and vegetation over two 3-band sets (B234, equivalent to SPOT)and B345.

Competitive learning networks for unsupervised training

Abstract Unsupervised training can play an important role in a hybrid classification system. Many clustering techniques such as K-means have been employed in unsupervised training. In this study competitive learning networks are proposed as unsupervised training methods. The Jeffries-Matusita (J-M) distance, which is a measure of statistical separability of pairs of the ‘trained’ classes, was used to evaluate the capability of the proposed methods. The simulation results and comparisons with K-means are provided.