Segmented canonical discriminant analysis of in situ hyperspectral data for identifying 13 urban tree species

A total of 458 in situ hyperspectral data were collected from 13 urban tree species in the City of Tampa, FL, USA using a spectrometer. The 13 species include 11 broadleaf and two conifer species. Three different techniques, segmented canonical discriminant analysis (CDA), segmented principal component analysis (PCA) and segmented stepwise discriminate analysis (SDA), were applied and compared for dimension reduction and feature extraction. With each of the three techniques, 10 features were extracted or selected from four spectral regions, visible (VIS: 1412–1797 nm), near-infrared (NIR: 707–1352 nm), mid-infrared 1 (MIR1: 1412–1797 nm) and mid-infrared 2 (MIR2: 1942–2400 nm), and used to discriminate the 13 urban tree species with a linear discriminate analysis (LDA) method. The cross-validation results, based on training samples that were used in the feature reduction step, and the results calculated from the test samples were used for evaluating the ability of the in situ hyperspectral data and performance of the segmented CDA, PCA and SDA to identify the 13 tree species. The experimental results indicate that a satisfactory discrimination of the 13 tree species was achieved using the segmented CDA technique (average accuracy (AA) = 96%, overall accuracy (OAA) = 96% and kappa = 0.958 from the cross-validation results; AA = 90%, OAA = 90% and kappa = 0.896 from the test samples) compared to the segmented PCA and SDA techniques, respectively (AA = 76% and 86%, OAA = 78% and 87%, and kappa = 0.763 and 0.857 from the cross-validation results; AA = 79% and 88%, OAA = 80% and 89%, and kappa = 0.782 and 0.879 from the test samples). In this study, the segmented CDA transformation is effective for dimension reduction and feature extraction for species discrimination with a relatively limited number of training samples. It outperformed the segmented PCA and SDA methods and produced the highest accuracies. The NIR and MIR1 regions have greater power for identifying the 13 species compared to the VIS and MIR2 spectral regions. The results indicate that CDA or segmented CDA could be applied broadly in mapping forest cover types, species identification and/or other land use/land cover classification practices with hyperspectral remote sensing data.     

Face recognition in hyperspectral images

Hyperspectral cameras provide useful discriminants for human face recognition that cannot be obtained by other imaging methods. We examine the utility of using near-infrared hyperspectral images for the recognition of faces over a database of 200 subjects. The hyperspectral images were collected using a CCD camera equipped with a liquid crystal tunable filter to provide 31 bands over the near-infrared (0.7 /spl mu/m-1.0 /spl mu/m). Spectral measurements over the near-infrared allow the sensing of subsurface tissue structure which is significantly different from person to person, but relatively stable over time. The local spectral properties of human tissue are nearly invariant to face orientation and expression which allows hyperspectral discriminants to be used for recognition over a large range of poses and expressions. We describe a face recognition algorithm that exploits spectral measurements for multiple facial tissue types. We demonstrate experimentally that this algorithm can be used to recognize faces over time in the presence of changes in facial pose and expression.     

Using vegetation reflectance variability for species level classification of hyperspectral data

A study was conducted to investigate whether reflectance data from vegetation in a tropical forest canopy could be used for species level discrimination. Reflectance spectra of 11 species were analysed at the scale of the leaf, branch, tree and species. To enhance separation of species-of-interest spectra from the other spectra in the data, the variation in reflectance values for the species-of-interest were used to create a characteristic spectral shape. With a simple algorithm, the resultant shape-space was used as a data filter that correctly discriminated against 94% of the non-species-of-interest trees.     

Conifer species recognition: Effects of data transformation

In situ hyperspectral data obtained with a high spectral resolution radiometer were analysed for identification of six conifer species. Hyperspectral data were measured in the summer and late fall seasons at 15-20 cm above portions of tree canopies from both the sunlit and shaded sides. An artificial neural network algorithm was applied for identification purposes. Six types of transformation were applied to the hyperspectral reflectance data ( R ), preprocessed with a simple smoothing, followed by band aggregation. These include log( R ), first derivative of R, first derivative of log( R ), normalized R, first derivative of normalized R, and log(normalized R ). First derivative of log( R ) and first derivative of normalized R resulted in best species recognition accuracies with greater than 90% average accuracies, more than 20% greater than the average accuracy obtained from the pre-processed hyperspectral data. The effect of hyperspectral data taken from the shade sides of tree canopies can be minimized by applying normalization or by taking the derivatives after applying a logarithm to the pre-processed data. We found that a big difference in solar angle did not cause a noticeable difference in accuracies of species recognition.     

Object-based fusion for urban tree species classification from hyperspectral,panchromatic and nDSM data

This study aims at identifying the best object-based fusion strategy that takes advantage of the complementarity of several heterogeneous airborne data sources for improving the classification of 15 tree species in an urban area (Toulouse, France). The airborne data sources are: hyperspectral Visible Near-Infrared (160 spectral bands, spatial resolution of 0.4 m) and Short-Wavelength Infrared (256 spectral bands, 1.6 m), panchromatic (14 cm), and a normalized Digital Surface Model (12.5 cm). Object-based feature and decision level fusion strategies are proposed and compared when applied to a reference site where the species are previously identified during ground truth collection. This allows the best fusion strategy to be selected with a view to introducing the method in an automatic process (tree crown delineation and species classification) on a test site, independent of the reference site used for learning. In particular, a decision level fusion is selected: based on the Support Vector Machine algorithm, Visible Near-Infrared and Short-Wavelength Infrared classifications use Minimum Noise Fraction components at the original spatial resolution, whereas panchromatic and normalized Digital Surface Model classifications use, respectively, Haralick’s and structural features computed at the object scale. After the computation of a decision profile for each source at the object level based on the classification algorithms’ membership probabilities, these decision profiles are combined and a decision rule is applied to predict the species. Focusing on the reference site, the Visible Near-Infrared exhibits the best performances with F-score values higher than 60% for 13 species out of 15. The Short-Wavelength Infrared is the most powerful for three species with F-score greater than 60% for seven common species with the Visible Near-Infrared. The panchromatic and normalized Digital Surface Model contribute marginally. The best fusion strategy (decision fusion) does not improve significantly the overall accuracy with 77% (kappa = 74%) against 75% (kappa = 72%) for the Visible Near-Infrared but in general, it improves the results for cases where complementarities have been observed. When applied to the test site and assessed for the two majority species (Tilia tomentosa and Platanus x hispanica), the selected approach gives consistent results with an overall accuracy of 63% against 55% for the Visible Near-Infrared.     

Classification of coniferous tree species and age classes using hyperspectral data and geostatistical methods

Classifications of coniferous forest stands regarding tree species and age classes were performed using hyperspectral remote sensing data (HyMap) of a forest in western Germany. Spectral angle mapper (SAM) and maximum likelihood (ML) classifications were used to classify the images. Classification was performed using (i) spectral information alone, (ii) spectral information and stem density, (iii) spectral and textural information, (iv) all data together, and results were compared. Geostatistical and grey level co‐occurrence matrix based texture channels were derived from the HyMap data. Variograms, cross variograms, pseudo‐cross variograms, madograms, and pseudo‐cross madograms were tested as geostatistical texture measures. Pseudo‐cross madograms, a newly introduced geostatistical texture measure, performed best. The classification accuracy (kappa) using hyperspectral data alone was 0.66. Application of pseudo‐cross madograms increased it to 0.74, a result comparable to that obtained with stem density information derived from high spatial resolution imagery.     

A comprehensive review of hyperspectral data fusion with lidar and sar data

With the development of remote sensing techniques, the fusion of multimodal data, particularly hyperspectral-Light Detection And Ranging (HS-LiDAR) and hyperspectral-SAR, has become an important research field in numerous application areas. Multispectral, HS, LiDAR, and Synthetic Aperture Radar (SAR) images contain detailed information about the monitored surface that are complementary to each other. Thus, data fusion methods have become a promising solution to obtain high spatial resolution remote-sensing images. The main point of this review paper is to classify hyperspectral-LiDAR and hyperspectral-SAR data fusion with approaches. Moreover, recent achievements in the fusion of hyperspectral-LiDAR and hyperspectral-SAR data are highlighted in terms of faced challenges and applications. Most frequently used data fusion datasets that include IEEE GRSS Data Fusion Contests are also described.     

Discrimination of lianas and trees with leaf-level hyperspectral data

Lianas are an important component of the biological diversity in two tropical forests with contrasting moisture regimes in Panama. However, their presence in a tree crown may be a source of confusion in remotely sensed data collected for inventories or assessment of vegetation health. The structural growth form of lianas contrasts with trees in that their proportion of leafy biomass to woody biomass is much higher. In effect, they use trees for structural support and typically form a monolayer of leaves above the crown of the supporting tree. Here, we investigated possible differences between hyperspectral signatures of lianas and trees at the leaf level using pattern recognition techniques. Our method involves principal components analysis followed by training and classification using a selection of supervised parametric and nonparametric classifiers. At a tropical dry forest site (Parque Natural Metropolitano), lianas and trees are distinguishable as groups based on their leaf spectral reflectance characteristics in dry season conditions. Classification was improved using ancillary data on leaf chlorophyll content. Their distinction at this site may be related to drought stress and/or phenological differences between the two groups. At a tropical wet forest site (Fort Sherman), discrimination between the two groups was not as clear. Additional research is required to determine the physiological basis of possible differences as well as to determine if these differences are observable at the canopy level.     

Improving within-genus tree species discrimination using the discrete wavelet transform applied to airborne hyperspectral data

Discrete wavelet analysis was assessed for its utility in aiding discrimination of three pine species (Pinus spp.) using airborne hyperspectral data (AVIRIS). Two different sets of Haar wavelet features were compared to each other and to calibrated radiance, as follows: (1) all combinations of detail and final level approximation coefficients and (2) wavelet energy features rather than individual coefficients. We applied stepwise discriminant techniques to reduce data dimensionality, followed by discriminant techniques to determine separability. Leave-one-out cross validation was used to measure the classification accuracy. The most accurate (74.2%) classification used all combinations of detail and approximation coefficients, followed by the original radiance (66.7%) and wavelet energy features (55.1%). These results indicate that application of the discrete wavelet transform can improve species discrimination within the Pinus genus.     

Distinguishing mangrove species with laboratory measurements of hyperspectral leaf reflectance

As a first step in developing classification procedures for remotely acquired hyperspectral mapping of mangrove canopies, we conducted a laboratory study of mangrove leaf spectral reflectance at a study site on the Caribbean coast of Panama, where the mangrove forest canopy is dominated by Avicennia germinans, Laguncularia racemosa, and Rhizophora mangle. Using a high‐resolution spectrometer, we measured the reflectance of leaves collected from replicate trees of three mangrove species growing in productive and physiologically stressful habitats. The reflectance data were analysed in the following ways. First, a one‐way ANOVA was performed to identify bands that exhibited significant differences (P value<0.01) in the mean reflectance across tree species. The selected bands then formed the basis for a linear discriminant analysis (LDA) that classified the three types of mangrove leaves. The contribution of each narrow band to the classification was assessed by the absolute value of standardised coefficients associated with each discriminant function. Finally, to investigate the capability of hyperspectral data to diagnose the stress condition across the three mangrove species, four narrow band ratios (R ₆₉₅/R ₄₂₀, R ₆₀₅/R ₇₆₀, R ₆₉₅/R ₇₆₀, and R ₇₁₀/R ₇₆₀ where R ₆₉₅ represents reflectance at wavelength of 695nm, and so on) were calculated and compared between stressed and non‐stressed tree leaves using ANOVA.

Results indicate a good discrimination was achieved with an average kappa value of 0.9. Wavebands at 780, 790, 800, 1480, 1530, and 1550 nm were identified as the most useful bands for mangrove species classification. At least one of the four reflectance ratio indices proved useful in detecting stress associated with any of the three mangrove species. Overall, hyperspectral data appear to have great potential for discriminating mangrove canopies of differing species composition and for detecting stress in mangrove vegetation.     

Estimating aboveground biomass of grassland having a high canopy cover: an exploratory analysis of in situ hyperspectral data

To improve the estimation of aboveground biomass of grassland having a high canopy cover based on remotely sensed data, we measured in situ hyperspectral reflectance and the aboveground green biomass of 42 quadrats in an alpine meadow ecosystem on the Qinghai–Tibetan Plateau. We examined the relationship between aboveground green biomass and the spectral features of original reflectance, first-order derivative reflectance (FDR), and band-depth indices by partial least squares (PLS) regression, as well as the relationship between the aboveground biomass and narrow-band vegetation indices by linear and nonlinear regression analyses. The major findings are as follows. (1) The effective portions of spectra for estimating aboveground biomass of a high-cover meadow were within the red-edge and near infrared (NIR) regions. (2) The band-depth ratio (BDR) feature, using NIR region bands (760–950 nm) in combination with the red-edge bands, yields the best predictive accuracy (RMSE?=?40.0 g m^?2) for estimating biomass among all the spectral features used as independent variables in the partial least squares regression method. (3) The ratio vegetation index (RVI2) and the normalized difference vegetation index (NDVI2) proposed by Mutanga and Skidmore (Mutanga, O. and Skidmore, A.K., 2004a Mutanga, O. and Skidmore, A. K. 2004a. Narrow band vegetation indices solve the saturation problem in biomass estimation. International Journal of Remote Sensing, 25: 1–16.  [Google Scholar], Narrow band vegetation indices solve the saturation problem in biomass estimation. International Journal of Remote Sensing, 25, pp. 1–6) are better correlated to the aboveground biomass than other VIs (R ²?=?0.27 for NDVI2 and 0.26 for RVI2), while RDVI, TVI and MTV1 predicted biomass with higher accuracy (RMSE?=?37.2 g m^?2, 39.9 g m^?2 and 39.8 g m^?2, respectively). Although all of the models developed in this study are probably acceptable, the models developed in this study still have low accuracy, indicating the urgent need for further efforts.     

Investigating the spectral and ecological characteristics of grassland communities across an ecological gradient of the Inner Mongolian grasslands with in situ hyperspectral data

Grassland degradation is serious in the Mongolian plateau, especially in Inner Mongolia, China. Accurate monitoring of grassland types and qualities is increasingly important for the purposes of grassland conservation and restoration. Using in situ hyperspectral reflectance data and ground-based ecological measurements, we explored the potential for large-scale monitoring grassland communities using imaging spectroradiometers. We compared the spectral reflectance of the major types of grasslands and field plots with/without livestock grazing. We also did statistical analysis about the relationship between hyperspectral indices and aboveground biomass (AGB) of the surveyed grassland communities. The results showed that: (1) the dominant plant species varied across meadow, typical, and desert steppe, and they also varied between fenced and grazed plots; (2) in situ hyperspectral data are useful for differentiating grassland communities of meadow, typical, and desert steppe and grassland communities with and without livestock grazing; and (3) the prediction accuracies of vegetation indices for AGB decreased from desert to typical and meadow steppe, and the results were contrary for the prediction accuracies of red edge inflection point (REIP). REIP may not be suitable for estimating AGB of the low-density grassland communities. The above results implied that care must be taken while using statistical models to link spectral and ecological measurements in large geographical scales since there is lack of portability over different types of grassland communities. This study provides foundations for future large-scale efforts of monitoring grassland communities in Inner Mongolia using imaging spectroradiometers.     

Aliasing assessment in wavelength domain of hyperspectral data

This paper focuses on a procedure for the assessment of spectral aliasing in hyper-spectral data acquired by push-broom spectrometers. The procedure is based on a push-broom spectrometer model that simulates acquired spectra by taking into account only the instrument parameters; aliasing is measured by means of some figures of merit considered among those proposed in literature. Quantitative evaluations have been performed on simulated spectra both with and without ideal atmospheric and radiometric correction. Results are presented in this paper; the impact of spectrometer slit size on aliasing appearance is also addressed.     

Grass species differentiation through canopy hyperspectral reflectance

This study attempts (1) to evaluate the capability of hyperspectral reflectance to differentiate C₃ and C₄ grass species, both in isolation and in mixed canopies; (2) to identify the critical spectral ranges that differentiate the two groups and individual species within them; and (3) to determine if there is temporal variation in these capabilities. During one year, hyperspectral reflectance of C₃ and C₄ grass species was measured both in single-species and in mixed canopies. Spectral bands with higher differentiating potential were identified and species classified. For single-species canopies, hyperspectral reflectance differentiated the two functional groups and most species in all seasons. In mixed canopies, it underestimated the fractional cover of the C₄ component. The green, red, and near infrared above 820 nm spectral ranges were critical both for species and functional group differentiation. In conclusion, hyperspectral information was useful to differentiate pure canopies, but the differentiation algorithms were season-specific. Additionally, we need to improve our understanding of interactive effects of species in order to accurately estimate the composition of assemblages.     

Nitrogen detection with hyperspectral normalized ratio indices across multiple plant species

The main focus of recent studies relating vegetation leaf chemistry with remotely sensed data is the prediction of chlorophyll and nitrogen content using indices based on a combination of bands from the red and infrared wavelengths. The use of high spectral resolution data offers the opportunity to select the optimal wavebands for predicting plant chemical properties. In order to test the optimal band combinations for predicting nitrogen content, normalized ratio indices were calculated for all wavebands between 350 and 2200 nm for five different species. The correlation between these indices and the nitrogen content of the samples was calculated and compared between species. The results show a strong correlation between individual normalized ratio indices and the nitrogen content for different species. The spectral regions that are most effective for predicting nitrogen content are, for each individual species, different from the normalized difference vegetation index (NDVI) spectral region. By combining the areas of maximum correlation it was possible to determine the optimal spectral regions for predicting leaf nitrogen content across species. In a cross‐species situation, normalized ratio indices using the combination of reflectance at 1770 nm and at 693 nm may give the best relation to nitrogen content for individual species.     

Geothermal exploration with Hymap hyperspectral data at Brady-Desert Peak, Nevada

We explore the use of Hymap (hyperspectral mapper) to remotely map unique geothermal indicator minerals over the Brady-Desert Peak geothermal fields. Geothermal-related minerals and rocks such as sinter, tufa, and sulfates, display diagnostic characteristics in the visible and shortwave infrared; their presence and distribution can be used to guide more detailed field work for geothermal exploration. The Brady-Desert Peak geothermal fields are located about 80 km east of Reno, Nevada in the Hot Springs Mountains. North-northeast-striking en-echelon faults offset Tertiary volcanic and lacustrine rocks. Two geothermal power plants produce electricity from two separate geothermal systems, one with numerous fumaroles and mudpots, the other showing no active surface expression of geothermal activity. Surface occurrences of gypsum, calcium-carbonate, hematite, and opaline silica were identified at both sites with the hyperspectral data; these minerals when considered together are indicative of geothermal activity at both sites. Mapping results were synthesized with other spatial data in a geographic information systems (GIS) database that was used to help draw structural interpretations of faulting and fault controls at the Brady-Desert Peak area. The same processing methods can be applied to new hyperspectral data sets for future exploration in the Great Basin, especially in areas that lack obvious thermal expressions.     

Ridge regression-based feature extraction for hyperspectral data

Feature extraction based on ridge regression (FERR) is proposed in this article. In FERR, a feature vector is defined in each spectral band using the mean of all classes in that dimension. Then, it is modelled using a linear combination of its farthest neighbours from among other defined feature vectors. The representation coefficients obtained by solving the ridge regression model compose the projection matrix for feature extraction. FERR can extract each desired number of features while the other methods such as linear discriminant analysis (LDA) and generalized discriminant analysis (GDA) have limitations in the number of extracted features. Experimental results on four popular real hyperspectral images show that the efficiency of FERR is superior to those of other supervised feature extraction methods in small sample-size situations. For example, for the Indian Pines dataset, the comparison between the highest average classification accuracies achieved by different feature extraction methods using a support vector machine (SVM) classifier and 16 training samples per class shows that FERR is 7% more accurate than nonparametric weighted feature extraction and is also 9% better than GDA. LDA, having the singularity problem in the small sample-size situations, has 40% less accuracy than FERR. The experiments show that generally the performance of FERR using the SVM classifier is better than when using the maximum likelihood classifier.     

Daytime fire detection using airborne hyperspectral data

The shortwave infrared region of the electromagnetic spectrum, covering wavelengths from 1400 to 2500 nm, can include significant emitted radiance from fire. There have been relatively few evaluations of the utility of shortwave infrared remote sensing data, and in particular hyperspectral remote sensing data, for fire detection. We used an Airborne Visible InfraRed Imaging Spectrometer (AVIRIS) scene acquired over the 2003 Simi Fire to identify the hyperspectral index that was able to most accurately detect pixels containing fire. All AVIRIS band combinations were used to calculate normalized difference indices, and kappa was used to compare classification ability of these indices for three different fire temperature ranges. The most accurate index was named the Hyperspectral Fire Detection Index (HFDI). The HFDI uses shortwave infrared bands centered at 2061 and 2429 nm. These bands are sensitive to atmospheric attenuation, so the impacts of variable elevation, solar zenith angle, and atmospheric water vapor concentration on HFDI were assessed using radiative transfer modeling. While varying these conditions did affect HFDI values, relative differences between background HFDI and HFDI for 1% fire pixel coverage were maintained. HFDI is most appropriate for detection of flaming combustion, and may miss lower temperature smoldering combustion at low percent pixel coverage due to low emitted radiance in the shortwave infrared. HFDI, two previously proposed hyperspectral fire detection indices, and a broadband shortwave infrared-based fire detection index were applied to AVIRIS scenes acquired over the 2007 Zaca Fire and 2008 Indians Fire. A qualitative comparison of the indices demonstrated that HFDI provides improved detection of fire with less variability in background index values.     

基于一类支持向量机的高光谱影像地物识别

高光谱遥感影像具有丰富的光谱信息,在地物识别方面具有明显的优势。一类支持向量机(OCSVM)不仅保留了支持向量机的原有优势,而且只需要待识别类型的训练样本。为此提出了算法,通过数学模型选择、核函数设计与参数的自适应调整将OCSVM原理融入到高光谱影像的地物识别算法中,提高了识别的精度,降低了对训练样本的要求。最后利用两幅高光谱影像进行了实验分析,实验结果证明了所提算法的有效性。     

Some issues in the classification of DAIS hyperspectral data

Classification accuracy depends on a number of factors, of which the nature of the training samples, the number of bands used, the number of classes to be identified relative to the spatial resolution of the image and the properties of the classifier are the most important. This paper evaluates the effects of these factors on classification accuracy using a test area in La Mancha, Spain. High spectral and spatial resolution DAIS data were used to compare the performance of four classification procedures (maximum likelihood, neural network, support vector machines and decision tree). There was no evidence to support the view that classification accuracy inevitably declines as the data dimensionality increases. The support vector machine classifier performed well with all test data sets. The use of the orthogonal MNF transform resulted in a decline in classification accuracy. However, the decision‐tree approach to feature selection worked well. Small increases in classifier accuracy may be obtained using more sophisticated techniques, but it is suggested here that greater attention should be given to the collection of training and test data that represent the range of land surface variability at the spatial scale of the image.