Skip to content
BY-NC-ND 3.0 license Open Access Published by De Gruyter Open Access September 13, 2016

An interactive tool for semi-automatic feature extraction of hyperspectral data

  • Zoltán Kovács EMAIL logo and Szilárd Szabó
From the journal Open Geosciences

Abstract

The spectral reflectance of the surface provides valuable information about the environment, which can be used to identify objects (e.g. land cover classification) or to estimate quantities of substances (e.g. biomass). We aimed to develop an MS Excel add-in – Hyperspectral Data Analyst (HypDA) – for a multipurpose quantitative analysis of spectral data in VBA programming language. HypDA was designed to calculate spectral indices from spectral data with user defined formulas (in all possible combinations involving a maximum of 4 bands) and to find the best correlations between the quantitative attribute data of the same object. Different types of regression models reveal the relationships, and the best results are saved in a worksheet. Qualitative variables can also be involved in the analysis carried out with separability and hypothesis testing; i.e. to find the wavelengths responsible for separating data into predefined groups. HypDA can be used both with hyperspectral imagery and spectrometer measurements. This bivariate approach requires significantly fewer observations than popular multivariate methods; it can therefore be applied to a wide range of research areas.

Highlights

  1. Reflectance curves are closely connected to the characteristics of materials.

  2. An MS Excel add-in was developed to analyse spectral data.

  3. Calculations are conducted with user defined formulas, considering all wavelengths.

  4. Regression models explore the correlations between spectral and attribute data.

  5. Separability testing reveals the greatest differences in the spectra of data groups.

1 Introduction

The appearance of hyperspectral imaging in remote sensing technology was a milestone in both data acquisition and the possibilities for identification of different objects or materials. Hyperspectral bands are narrow (with an approximate maximum width of 10 nm), continuous, and cover the visible and near-infrared range. Beside aerial sensors, there are field equipment and laboratory devices capable of recording the spectral characteristics of given surfaces or substances [1]. Consequently, wider research communities started to use spectrum analysis in different fields, from remote sensing to analytical chemistry or physics [27].

The technology does not require any preparation of the analysed substance (which can be expensive and requires a laboratory background with specific devices) and is non-destructive. Although, the acquired result is not the same as results achieved with traditional analytical methods; it is a reflectance curve having certain sections (e.g. minimums, maximums, peaks or slopes) indicating the presence or even the quantity of the analysed substance. Accordingly, we need to find those bands or their combinations, which can substitute laboratory analyses.

Using spectral bands, we can determine spectral indices which have a high correlation with a measurable quantity of surface objects. The best known index is the NDVI (Normalized Difference Vegetation Index [8]), which is widely used among researchers. Besides, there are many indices reported in several papers which aim, for example, to reduce the effect of soils’ reflectance on vegetation indices (SAVI, TSAVI, PVI) [9, 10], to identify water (NDWI) [11] and to determine the water content of vegetation [12] or its chlorophyll content [13]. Finding the best indices becomes complicated as increasing number of bands are involved, i.e. hyperspectral data allows us to find specific details; however, the possible band combinations increase exponentially.

Parallel with the larger dataset, the difficulty of data processing also increases. Given the large number of bands (generally > 100), several researchers have used multivariate techniques (usually Principal Component Analysis, PCA) to reduce the redundancy with uncorrelated principal components (PCs), and at the same time to maintain the maximum explained variance [14]. Although PCA is a very effective method in data processing, it has its limitations in terms of the minimum number of cases: a dataset is required which is 5–10 times larger than the number of variables (i.e. bands [15]), otherwise calculations are not accurate. If the aim is a prediction, PCR (Principal Component Regression) can be performed [16] and if we have large number of variables (i.e. bands) from the spectral analysis, the statistical analysis can only be efficient if we perform lots of chemical analysis, too. As the chemical (and all kinds of laboratory) analysis is expensive, the case number is limited, i.e. in some cases PCA cannot be performed or provides false results. Thus, the number of bands has to be reduced by the researcher, either by visual selection based on the spectral curve or by merging neighbouring bands (and calculating mean values). Besides, Jimenez and Landgrebe (1999) [17] criticized PCA, because it does not take into consideration the statistical properties of the variables. As another multivariate approach, Kooistra et al. (2001) [18] applied the partial least squares regression method to filter out latent factors from the bands, so they can efficiently predict the dependent variable.

There are several software packages providing statistical analysis, but none of them is able to prepare the spectral dataset for further investigations. For data preparation and data processing we developed a Visual Basic MS Excel add-in called Hyperspectral Data Analyst (HypDA). Excel is one of the most widely used software programs and its capabilities can be developed with small-sized add-ins with a user-friendly interface. HypDA is able to import a wide scale of input data from different devices, to handle the measurement replications, and to conduct batch processing of spectral data.

The aim of this development was to prepare a tool being able to process the spectral curves along the following driving forces: (1) to find the strongest relationship between the shape of the spectra and the observed attributes (the independent variable is scale data) or (2) to find the largest difference between groups based on the shape of the spectra (the independent variable is nominal data); furthermore, (3) its requirement for the case number is minimal; and, (4) the results can be used to interpolate the laboratory measurements with the help of the spectral curves.

2 Workflow

The concept of the data processing is summarized on Figure 1. All hyperspectral image processing software (e.g. ENVI, Erdas, Idrisi) is able to export the values of marked pixels of aerial or satellite images into ASCII files (e.g. ENVI ROI ASCII text files); furthermore, most field/laboratory spectrometers and spectroradiometers can export the data to MS Excel file format. Due to the developed flexible MS Excel environment we can easily convert these wide scale of input data to tables (i.e. worksheets) from different devices. HypDA needs a special organization of the data: the first worksheet should contain the spectral data of samples as independent variables, while the second contains qualitative data (as grouping variables: sampling sites, land use categories, soil types etc.) and/or quantitative data (e.g. pH, CaCO3 content, ion concentrations etc.) as dependent variables.

Figure 1 Theoretical workflow.
Figure 1

Theoretical workflow.

In the data pre-processing phase we may want to reduce the spectral noise, and there are several techniques to perform it, such as Savizky-Golay smoothing [19], normalization, standard normal variate (SNV) transformation [20], multiplicative scatter correction (MSC) [21], and first and second derivatives (using Savizky-Golay convolution coefficients) [22]. Spectral data can be cut with minimum and/or maximum wavelengths and can be placed in spectral ranges with user-defined widths to merge intensity values and replace them with a single calculated (mean or median) value. HypDA can also handle replications of measurements, calculating mean or median values using the similarity of variable names.

We can conduct quick tests with the workbook and its worksheet of spectral intensity values, and the worksheet of nominal and/or scale properties: visual interpretation of diagrams (scatterplot or boxplot), calculation of e.g. Jeffries-Matusita distance (J-M) [23] or Bhattacharyya distance between selected groups. A deeper analysis can be performed to reveal the strongest relationship between the shape of spectra (independent variable) and scale type attributes (dependent variable) is calculated with cross-validated regression models. The equation of these linear or non-linear fitted curves can be used to estimate the dependent variables. In addition, we can obtain spectral indices for the better discrimination of predefined groups compared the efficiency of original bands: the largest difference between user-defined groups is determined with the separability (J-M) or hypothesis testing. In this latter case, all groups have unique equations that are used for the classification of hyperspectral images, in hyperspectral image processing software.

3 Description of HypDA

3.1 HypDA structure

When the HypDA add-in is enabled, a new ribbon tab will be added to the Excel ribbon. A wizard was designed to break down the creation of a new HypDA-compatible workbook into multiple steps, from the selection of the input data to the customization of the output file. This new Excel workbook is required for further investigations. Workbooks have six worksheets (WS) with specific functions.

The first WS (‘spectral’) contains the independent values (spectral data), and the second WS (‘properties’) contains the grouping and/or dependent variables (physical and chemical attributes). In order to execute the calculation only on a part of the spectra (e.g. to exclude certain spectral ranges), or on a part of the samples (e.g. to exclude some sample group), bands and samples can be enabled or disabled before investigations. New properties can be added by reclassifying an existing attribute or calculating from other quantitative attributes (e.g. NDVI). The third WS (‘analysis’) is used to observe a selected model in greater detail, while the fourth WS (‘table’) is used to calculate all existing combinations of chosen bands in a matrix-like table. The fifth WS (‘best’) contains the main parameters of the best models from the fourth WS. The sixth WS (‘settings’) is hidden from the user and stores all the saved parameters and user settings.

3.2 Working with parametric formulas

HypDA tool is able to define parametric formulas referencing on spectral intensity values (e.g. BandA/BandB) and to use the results obtained as dependent variables. Users can follow two different procedures: (1) to choose certain bands (e.g. BandA = 560 nm and BandB = 970 nm) that result in a certain single model (on ‘analysis’ WS); (2) or the other possibility is to involve all bands, and to generate a matrix-like table with all the possible pairs of bands (on ‘table’ WS), where the bands change column by column (BandA) and row by row (BandB). In this second case an n-element formula results in an n-dimension table, so if a three-element formula is defined – e.g. BandC/(BandA-BandB) – the result is a three dimensional table; consequently BandC acquires new intensity values following the calculations of each 2D table. Finally, a four-element formula results in a four-dimensional table where BandD acquires new intensity values after the calculations of each 3D table. Necessarily, this solution implies that all former 2D tables have to be overwritten. To avoid the loss of the data gained, a user-defined number (up to 100) of the best values is saved to the fifth (‘best’) WS from each table together with all the parameters which are needed to reconstruct the models later.

This second approach means that if the ‘spectral’ WS has 100 bands, a table will contain 10,000 individual models, and if a four-element formula is used, 10,000 tables with 10,000 individual models, i.e. 100 million regression models are calculated, and the best 10,000 models are saved for further investigations.

When we use matrix-like tables to search for high correlations or large differences, all items of the table constitute a unique investigated model described by a single value that represents how good the model is (e.g. R2). One can choose a cell from the table or the ‘best’ WS which contains a model in the background and recall this model in the ‘analysis’ WS for a detailed investigation (Figure 2).

Figure 2 Any item of tables or any saved best value can be recalled in the ‘analysis’ WS.
Figure 2

Any item of tables or any saved best value can be recalled in the ‘analysis’ WS.

3.3 Data processing: examples of hyperspectral image analysis and results

Generally, when conducting data analysis, there are two main issues which arise during the processing of remotely sensed data: (1) which part of the spectra shows the strongest relationship between a certain phenomenon or a property, or (2) which part of the spectra shows the greatest difference between groups. Furthermore, (3) we can visualize and query the matrix of the correlations, R2 or RMSE values by band combinations. The following case studies represent the capabilities of HypDA with data collected by a spectrometer and with a hyperspectral image.

3.3.1 Searching for the bands which best correlate with the attribute data

Our main goal was to investigate the relationship between the values of chosen attributes and the calculated values from the intensity values of given bands. Following this, regression models (defined by the user: one at a time, or even all of the types simultaneously; linear, exponential, logarithmic, power and polynomial 2-4) are calculated between a chosen attribute (dependent variable) and the indices (independent variables).

All details of the model are listed by curve type (linear, exponential, logarithmic, power and polynomial 2-4), including the model error, predicted residual sum of squares (PRESS), coefficient of determination, residuals and regression sum of squares, adjusted R square, standard error for the y estimate, significance, quartiles of residuals, root-mean-square deviation and normalized root-mean-square deviation calculated for all available fitting curve types. Furthermore, a scatterplot with the data pairs and a fitted curve allows the possibility of visual analysis. A boxplot of the residuals and a homoscedasticity chart with standardized predictive and residual values, and influential values (using Cook’s distance), are calculated and plotted for the selected fitting curve(s). If one sets a grouping (nominal) attribute, the data points of charts can be coloured separately and the coefficient of determinations can be calculated for each group individually according to the chosen curve fitting type.

There are several methods embedded for the validation of the regression models: leave-one-out, k-fold and repeated random sub-sampling cross-validation is available. Datasets can be divided automatically by a user-defined ratio into a train and test dataset.

In order to show the relevance of the tool and to reveal how soil characteristics can be predicted by using spectral data we analysed 44 soil samples collected in the Bükk Mountains (N-Hungary), which physical and chemical characteristics (pH, CaCO3-content, organic matter, granulometrical composition) were determined in the Laboratory of Geosciences, University of Debrecen. Spectral features were measured with an Avantes 2048 spectrometer in the 200–1100 nm range, with the band with of 0.55 nm. CaCO3 was the dependent and the 1630 bands were the independent variables. All measurements were arranged into the required format with an import module of HypDA, then regression analyses were conducted involving all soil characteristics as independent variables and all bands as dependent variables (it is possible to apply equations for calculating spectral indices and to run the analyses in batch mode). Considering all bands, we identified a band (289 nm) that had high correlation with CaCO3 content (r = 0.92, p < 0.05). Furthermore, R2 was 0.85 (p < 0.001), and RMSE = 4.51 (the error compared to the average CaCO3 content was 16.23%) (Figure 3). The results were validated by splitting the dataset and recalculating the equation. Boostrapping with random selection of cases (100 times) resulted the lower and upper 95% confidence interval of the R2 between 0.840 and 0.857.

Figure 3 Main panels of the ‘Analysis’ WS for testing the correlations between spectral indices and attributes. Worksheet contains the data table (A) with the relevant information about samples, a scatter plot (B), a boxplot (C) and a plot for the homoscedasticity (D).
Figure 3

Main panels of the ‘Analysis’ WS for testing the correlations between spectral indices and attributes. Worksheet contains the data table (A) with the relevant information about samples, a scatter plot (B), a boxplot (C) and a plot for the homoscedasticity (D).

3.3.2 Searching for bands where user defined groups have significant differences

The tool is able to identify the largest difference between sample groups; accordingly, separability and/or hypotesis testing can be performed. In particular, both parametric and nonparametric hypothesis tests (the Mann-Whitney U-test, the Kruskal-Wallis H-test and the Welch t-test) can be carried out. As a first step, a normality test should be conducted (Shapiro-Wilk test) to decide whether parametric or non-parametric methods can be applied. Descriptive statistics (e.g. mean, percentiles, skewness and kurtosis) are also calculated for each group. The Welch t-test and the Mann-Whitney U-test are performed for each available group pairs and the Kruskal-Wallis H-test for all selected groups. Moreover, the Bonferroni correction (presented in tabular form) helps to decide whether there is a significant difference between the group pairs originating from the significance of Mann-Whitney tests [24]. Finally, effect sizes [25] are also calculated to quantify differences in a standardized and comparable way and a boxplot diagram is constructed to displays the quartiles of the groups.

Separability indices (such as Bhattacharyya distance, Jeffries – Matusita distance, divergence and transformed divergence) can be determined from the covariance matrices and the mean vectors of the analysed populations.

In order to find band combinations where a certain group can be absolutely distinguished from the others, three more analyses can be carried out: (1) have no overlapping inter-percentile ranges; (2) show a high Jeffries – Matusita distance and/or (3) indicate a significant Mann-Whitney test.

We demonstrate a case study to present the ability of our tool to discriminate land use/land cover (LULC) categories of a hyperspectral image captured by an AISA Eagle II sensor in the campus of University Debrecen (the city is the second largest city in Hungary) in the range of 400–1000 nm (128 bands). Seven LULC categories were analysed (asphalt, building, forest, grassland, synthetic grass, tennis court and shadow); we collected 100 points for the train and 1000 points for the ground truth datasets by LULC classes in ENVI (Exelis Visual Solutions, 2014) according to Burai et al. (2015) [26]. Spectral band data were exported as a ROI file and processed in MS Excel with the HypDA add-in. The next step was to apply the general equation of (BandA-BandB)/(BandA+BandB) to calculate spectral indices and chose the one which provided the best J-M values (where the number of LULC pairs was maximal in terms of the J-M distances being above a critical value defined by the user, e.g. 1.85). Results showed that almost all LULC classes can be discriminated with the help of the new indices, except the artificial group of classes (asphalt and buildings) (Figure 4). Results can be refined with the calculation of spectral indices per category when one category’s difference from the others is enhanced. In this case, the result was seven spectral indices for the seven classes (Table 1), which can be used in the classification process in ENVI (or any other image processing software).

Figure 4 Main panels of the ‘Analysis’ WS for testing differences between groups. Worksheet contains the data table (A) with the relevant information about pixels, a boxplot (B) for visualising the distribution of populations and cross-tables (C) with the selected hypothesis test or separability index.
Figure 4

Main panels of the ‘Analysis’ WS for testing differences between groups. Worksheet contains the data table (A) with the relevant information about pixels, a boxplot (B) for visualising the distribution of populations and cross-tables (C) with the selected hypothesis test or separability index.

Table 1

Spectral indices generated with the HypDA add-in for the discrimination of seven land use/land cover categories (each index were determined by categories with the enhancing the differences compared to all other categories).

Class

Index

Asphalt

744 nm / 615 nm

Building

682 nm / 672 nm

Forest

753 nm / 720 nm

TennisCourt

686 nm / 601 nm

SyntheticGrass

840 nm / 787 nm

Grass

517 nm / 485 nm

Shadow

522 nm / 445 nm

Finally, applying these indices, we conducted a Support Vector Machine (SVM) classification in ENVI and determined the accuracy of the outcomes. Comparing the solution with the original bands, HypDA based indices performed better (Figure 5). Based on the overall accuracy or Kappa Coefficient, the difference is slight but Producers’s Accuracy (PA) and User’s Accuracy (UA) calculated by categories (Table 2) highlight the advantage of HypDA indices: all classification was more successful than with the original bands, only in case of buildings and asphalt surfaces resulted smaller accuracy values (but PA or UA was higher in these cases, too).

Figure 5 Supervised classification of a hyperspectral image of Debrecen (University Campus), Hungary (A: the true colour orthophoto of the area; B: classification with the HypDA indices; C: classification with all original bands).
Figure 5

Supervised classification of a hyperspectral image of Debrecen (University Campus), Hungary (A: the true colour orthophoto of the area; B: classification with the HypDA indices; C: classification with all original bands).

Table 2

Accuracy assessment of the SVM classification considering the one performed with all original spectral bands and the one with the HypDA indices (larger accuracies were highlighted with bold letters).

Class

SVM

HypDA indices

PA (%)

UA (%)

PA (%)

UA (%)

Asphalt

96.24

74.71

93.42

80.91

Building

81.03

99.44

88.15

96.29

Forest

93.16

99.13

97.06

99.47

TennisCourt

99.21

98.00

99.50

99.67

SyntheticGrass

99.21

99.50

99.82

99.63

Grass

99.29

96.09

99.18

98.29

Shadow

99.60

88.64

99.41

94.27

3.3.3 Searching for the most frequent band combinations

When we consider all possible band pairs to find the best correlations or models having the lowest RMSE, the matrix can be plotted on a correlogram (i.e. all bands in rows vs all columns). Correlogram indicates the required parameters (e.g. R2, separability indices) with colours, based on user-defined formats. This operation enhances those spectra ranges which are generally stronger or weaker for the observed relations (Figure 6). When there is a large number of saved models on the ‘best’ WS, then a table can be created, showing which band pairs appear the most frequently as best values for specific formulas or relations. Figure 6 shows how a table can be coloured, based on user-defined condition formats. This option can be very helpful in creating our own hyperspectral indices based on our dataset on a given subject.

Figure 6 Coloured table emphasising the weaker and stronger relationships between selected attributes and calculated spectral indices.
Figure 6

Coloured table emphasising the weaker and stronger relationships between selected attributes and calculated spectral indices.

4 Results and Discussion

Hyperspectral datasets represent an important base for data mining. The analysis requires spectral bands and measured or observed values as independent variables. The number of hyperspectral bands is usually large, so an automated procedure is desirable. Multivariate techniques require large numbers of cases; however, expensive analyses of the examined materials in the laboratories can limit the number of measurements. Our approach considers all bands and we can use any kind of equations involving a maximum of 4 bands at the same time. According to the bivariate technique, it requires fewer cases.

A similar application was developed by Buddenbaum and Püschel (2012) [27] in EnMap-Box software [28] and the approach was successfully applied in a study by Buchorn et al. [29] using field spectroscopy. Beside the similarity (as HypDA also determines indices based on regression between a measured variable and the spectral bands), HypDA provides several methods to investigate the prerequisites of regression analysis. Users can verify homoscedasticity and can exclude the influential data points to ensure the normal distribution of the residuals (differences of measured and predicted values) automatically (based on Cook’s distance [15]) or interactively (based on the scatterplot of the variables with highlighted outliers). Any kind of tabular data can be processed and users can define any kind of equations (even with four bands); furthermore, calculations can be run in batch mode with several tasks.

Moreover, HypDA can determine those bands/band combinations, which can discriminate user-defined groups. Due to the matrix approach, all possible band combinations are taken into consideration and the best solutions can be chosen by the users with the help of several embedded possibilities (separability or hypothesis testing). We successfully applied the indices generated by HypDA in a land cover classification of Landsat data and gained 98% overall accuracy in discriminating five land cover classes [30].

A future task is to connect the add-in with a GIS software where the indices can be used directly with remotely sensed images. Another promising way of development is the integration to R software (R Core Team).

5 Conclusions

Spectral indices are efficient tools in estimating characteristics of materials or Earth’s surface and can be used in image classification. We developed a software add-in being able to determine spectral indices for both purposes. According to its usage, we found that spectral indices produced directly to given task can perform well. CaCO3 content of soils (Cambisols) was predicted with 16.3% relative error using the data of a spectrometer. Classification of a hyperspectral aerial image using spectral indices being determined to enhance the differences between seven LULC classes resulted a better overall accuracy than the one performed with the original bands.

6 HypDA availability and hardware requirements

HypDA is a free add-in and can be downloaded from its webpage after registration: https://sites.google.com/site/hyperspectraldataanalyst/download.

There is no special hardware requirement to use the add-in. Of course, the processing time is heavily dependent on the computer’s characteristics and the size of datasets. To indicate the calculation time on a standard PC (4 GB RAM, 3.3 GHz, OS: Windows 8.1 64 bit), a table with 10,000 Kruskal-Wallis tests on 1000 samples (divided into 10 groups) takes only 20 seconds.


Zoltán Kovács: Department of Physical Geography and Geoinformation Systems, University of Debrecen, Debrecen, Hungary Tel.: +36 52 512900/22201 (switchboard); Fax: +36 52 512945

Acknowledgement

Zoltán Kovács was supported by the European Union and the State of Hungary, co-financed by the European Social Fund in the framework of TÁ-MOP 4.2.4. A/2-11-1-2012-0001 ‘National Excellence Program’. The publication was supported by the University of Debrecen (RH/751/2015) and the SROP-4.2.2.B-15/1/KONV-2015-0001 project. The project has been supported by the European Union, co-financed by the European Social Fund.

References

[1] Chang, C 2003. Hyperspectral Imaging. Springer Science and Business Media, ISBN 978-1-4419-9170-610.1007/978-1-4419-9170-6Search in Google Scholar

[2] Ben-Dor, E., Chabrillat, S., Dematte, J.A.M., Taylor, G.R., Hill, J., Whiting, M.L., Sommer, S., 2009. Using imaging spectroscopy to study soil properties. Remote Sensing Environ. 113, S38-S55. 10.1016/j.rse.2008.09.019Search in Google Scholar

[3] Deng, C., Wu, C., 2012. BCI: A biophysical composition index for remote sensing of urban environments. Remote Sensing Environ. 127, 247–259. 10.1016/j.rse.2012.09.009Search in Google Scholar

[4] Fodor, N., Pásztor, L., Németh, T., 2014. Coupling the 4M crop model with national geo-databases for assessing the effects of climate change on agro-ecological characteristics of Hungary. Int. J. Digital Earth 7 (5), 391–410. 10.1080/17538947.2012.689998Search in Google Scholar

[5] Smith, A.M.S., Kolden, C.A., Tinkham, W.T., Talhelm, A.F., Marshall, J.D. et al., 2014. Remote sensing the vulnerability of vegetation in natural terrestrial ecosystems. Remote Sensing Environ. 154, 322–337. 10.1016/j.rse.2014.03.038Search in Google Scholar

[6] Szalai, Z., Kiss, K., Jakab, G., Sipos, P., Belucz, B., Németh, T., 2013. The use of UV-VIS-NIR reflectance spectroscopy to identify iron minerals. Astron. Nachr. 334, 940–943. 10.1002/asna.201211965Search in Google Scholar

[7] Thenkabail, P.S., Lyon, J.G., Huete, A., 2011. Hyperspectral remote sensing of vegetation. CRC Press- Taylor and Francis group.10.1201/b11222-41Search in Google Scholar

[8] Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1973. Monitoring vegetation systems in the great plains with ERTS. Third ERTS Symposium, NASA SP-351, 309–317.Search in Google Scholar

[9] Huete, A.R., 1988. A soil adjusted vegetation index (SAVI). Remote Sensing Environ. 25, 295–309. 10.1016/0034-4257(88)90106-XSearch in Google Scholar

[10] Richardson, A.J., Everitt, J.H., 1992. Using spectral vegetation indices to estimate rangeland productivity. Geocarto Int. 7(1), 63-69. 10.1080/10106049209354353Search in Google Scholar

[11] Gao, B.C., 1996. NDWI – A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sensing Environ. 58, 257–266. 10.1016/S0034-4257(96)00067-3Search in Google Scholar

[12] Ceccato, P., Gobron, N., Flasse, S., Pinty, B., Tarantola, S., 2002. Designing a spectral index to estimate vegetation water content from remote sensing data: Part 1: Theoretical approach. Remote Sensing Environ. 82, 188–197. 10.1016/S0034-4257(02)00037-8Search in Google Scholar

[13] Gitelson, A.A., Merzylak, M.N., 1997. Remote estimation of chlorophyll content in higher plant leaves. Int. J. Remote Sensing 18, 2691–2697. 10.1080/014311697217558Search in Google Scholar

[14] Leone, A.P., Sommer, S., 2000. Multivariate analysis of laboratory spectra for the assessment of soil development and soil degradation in the southern Apennines (Italy). Remote Sensing Environ. 72, 346–359. 10.1016/S0034-4257(99)00110-8Search in Google Scholar

[15] Kabacoff, R.I., 2011. R in action, Manning, Shelter IslandSearch in Google Scholar

[16] Jolliffe, I.T. 1982. A note on the Use of Principal Components in Regression. J. Royal Stat. Soc. Series C 31, 300–303. 10.2307/2348005Search in Google Scholar

[17] Jimenez, L.O., Landgrebe, D.A., 1999. Hyperspectral data analysis and supervised feature reduction via projection pursuit. IEEE Trans. Geosci. Remote Sensing 37, 2653-2667. 10.1109/36.803413Search in Google Scholar

[18] Kooistra, L., Wehrens, R., Leuven, R.S.E.W., Buydens, L.M.C., 2001. Possibilities of visible-near infrared spectroscopy for the assessment of soil contamination in river floodplains. Analytica Chimica Acta 446, 97-105. 10.1016/S0003-2670(01)01265-XSearch in Google Scholar

[19] Savitzky, A., Golay, M.J.E. 1964. Smoothing and Differentiation of Data by Simplified Least Squares Procedures. Anal. Chem. 36, 1627–39. 10.1021/ac60214a047Search in Google Scholar

[20] Barnes, R. J., Dhanoa, M.S. and Lister, S.J., 1989: Standard Normal Variate Transformation and De-trending of Near-Infrared Diffuse Reflectance Spectra. Appl. Spectrosc. 43, 772–777. 10.1366/0003702894202201Search in Google Scholar

[21] Geladi P. and Dåbakk E., 1995: An overview of chemometrics applications in near infrared spectrometry. J. Near Infrared Spectrosc., 3, 119–132. 10.1255/jnirs.63Search in Google Scholar

[22] Gans, P., Gill, J. B. 1983. Examination of the Convolution Method for Numerical Smoothing and Differentiation of Spectroscopic Data in Theory and in Practice. Applied Spectrosc. 37, 515–520. 10.1366/0003702834634712Search in Google Scholar

[23] Richards, J.A., Jia, X. 2005. Remote Sensing Digital Image Analysis. Springer, Berlin-Heidelberg-New York, ISBN-10 3-540-25128-6Search in Google Scholar

[24] Dunnett, C.W., 1964. New tables for multiple comparisons with a control. Biometrics 20, 482–491.10.2307/2528490Search in Google Scholar

[25] Cohen, J., 1992. Statistical power analysis. Curr. Dir. Psychol. 1, 98–101. 10.1111/1467-8721.ep10768783Search in Google Scholar

[26] Burai, P., Deák, B., Valkó, O., Pinty, B., Tomor, T., 2015. Classification of Herbaceous Vegetation Using Airborne Hyperspectral Imagery. Remote Sens. 7(2), 2046–2066. 10.3390/rs70202046Search in Google Scholar

[27] Buddenbaum, H.; Püschel, P. SpInMine (Spectral Index Data Mining Tool): Manual for Application: SpInMine (1.0); University of Trier: Trier, Germany, 2012.Search in Google Scholar

[28] van der Linden, S., Rabe, A.; Held, M., Jakimow, B., Leitão, P.J., Okujeni, A., Schwieder, M., Suess, S., Hostert, P. 2015. The EnMAP-Box – A Toolbox and Application Programming Interface for EnMAP Data Processing. Remote Sens. 7, 11249–11266.10.3390/rs70911249Search in Google Scholar

[29] Buchhorn M, Walker D, Heim B, Raynolds M, Epstein H, Schwieder M. 2013. Ground Based Hyperspectral Characterization of Alaska Tundra Vegetation along Environmental Gradients. Remote Sens. 5, 3971–4005. 10.3390/rs5083971Search in Google Scholar

[30] Olasz, A., Kristóf, D., Belényesi, M., Bakos, K., Kovács, Z., Balázs, B., Szabó, Sz. 2015. IQPC 2015 Track: Water detection and classification on multi-resource remote sensing and terrain data. ISPRS Archives, XL-3/W310.5194/isprsarchives-XL-3-W3-583-2015Search in Google Scholar

Received: 2015-8-5
Accepted: 2016-1-29
Published Online: 2016-9-13
Published in Print: 2016-9-1

© 2016 Z. Kovács and S. Szabó, published by De Gruyter Open

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Downloaded on 2.3.2024 from https://www.degruyter.com/document/doi/10.1515/geo-2016-0040/html
Scroll to top button