Feature Mining for Hyperspectral Image Classification

Hyperspectral image provides abundant spectral information for remote discrimination of subtle differences in ground covers. However, the increasing spectral dimensions, as well as the information redundancy, make the analysis and interpretation of hyperspectral images a challenge. Feature extraction is a very important step for hyperspectral image processing. Feature extraction methods aim at reducing the dimension of data, while preserving as much information as possible. Particularly, nonlinear feature extraction methods (e.g. kernel minimum noise fraction (KMNF) transformation) have been reported to benefit many applications of hyperspectral remote sensing, due to their good preservation of high-order structures of the original data. However, conventional KMNF or its extensions have some limitations on noise fraction estimation during the feature extraction, and this leads to poor performances for post-applications. This paper proposes a novel nonlinear feature extraction method for hyperspectral images. Instead of estimating noise fraction by the nearest neighborhood information (within a sliding window), the proposed method explores the use of image segmentation. The approach benefits both noise fraction estimation and information preservation, and enables a significant improvement for classification. Experimental results on two real hyperspectral images demonstrate the efficiency of the proposed method. Compared to conventional KMNF, the improvements of the method on two hyperspectral image classification are 8 and 11%. This nonlinear feature extraction method can be also applied to other disciplines where high-dimensional data analysis is required.

Hyperspectral imaging is concerned with the measurement, analysis, and interpretation of spectra acquired from a given scene (or specific object) at a short, medium or long distance by an airbone or satellite sensor. Over the last few years, hyperspectral image data sets have been collected for a great amount of locations over the world, using a variety of instruments for Earth observation. Despite the increasing importance of hyperspectral images in remote sensing applications, there is no common repository of hyperspectral data intended to distribute and share hyperspectral data sets in the community. Quite opposite, the hyperspectral data sets which are available for public use are spread among different storage locations and present significant heterogeneity regarding the storage format, associated meta-data (if any), or ground-truth availability. As a result, the development of a standardized hyperspectral data repository is a highly desired goal in the remote sensing community. In this paper, we take a necessary first step towards the development of a digital repository for remotely sensed hyperspectral data. The proposed system allows uploading new hyperspectral data sets along with meta-data, ground-truth and analysis results, with the ultimate goal of sharing publicly available hyperspectral images within the remote sensing community. The database has been designed in order to allow storing relevant information for the hyperspectral data available through the system, including basic image characteristics (width, height, number of bands, format) and more advanced meta-data (ground-truth information, publications in which the data has been used). The current implementation consists of a front-end to ease the management of images through a web interface, thus containing both synthetic and real hyperspectral images from two highly representative instruments, such as NASAs Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS) over the Cuprite Mining District in Nevada. Most importantly, the developed system includes a spectral unmixing-based content based image retrieval (CBIR) functionality which allows searching for images on the spectral unmixing information (spectrally pure components or endmembers and their associated abundances in the scene). This information is stored as meta-data associated to each hyperspectral image instance, and then used to search and retrieve images based on information content. This paper presents the design of the system and a preliminary validation of the unmixing-based retrieval functionality using both synthetic and real hyperspectral images stored in the database.

Hyperspectral images contain information from a wider range of the electromagnetic spectrum than natural images which gives them potential for better classification ability. However, hyperspectral datasets are typically small due to the expensive equipment needed to obtain the images, which can limit classification performance. One solution to this problem is transfer learning, in which a model trained on one dataset is reused for a separate dataset. Research has shown that transfer learning between hyperspectral datasets can give improved performance over models without transfer learning when training data are limited. Since extra hyperspectral data are not always available, the solution proposed here is to instead use networks pretrained on natural image (i.e., red, blue, green, or RGB) datasets for transfer learning. By using various feature selection and feature extraction methods, extracted hyperspectral samples are transformed into a three-channel format to imitate an RGB image and are used for fine tuning the well-known ResNet, DenseNet, and VGG networks. Feature extraction methods include techniques like principal component analysis, which create lower dimensional features from high dimensional spectral data. Alternatively, feature selection methods aim to find the best set of existing channels to use for classification. Experimental results are obtained using two well-known hyperspectral datasets, showing 73.6% accuracy on Pavia University and 82.8% accuracy on Salinas with 25 training samples per class. Additional ensemble methods are implemented that utilize multiple networks and show an increase in accuracy of 4.4% and 3% for Pavia University and Salinas, respectively. These results demonstrate that networks pretrained on RGB datasets are suitable for transfer learning with hyperspectral image datasets and can achieve desirable performance given the proper preprocessing technique.

Determination of structural characteristics of forests at large scales is an important problem in both scientific studies and development of management practices. Light detection and ranging (LIDAR) waveform data have been demonstrated to be valuable for estimating forest structural parameters even in dense forests, although challenges inherent to the LIDAR acquisition systems must be addressed. A new approach for processing LIDAR waveform data to estimate forest structural parameters is proposed. It was applied to Laser Vegetation Imaging Sensor waveform data acquired over old-growth tropical forest in the La Selva Biological Station, Costa Rica. Linear and nonlinear feature extraction methods were utilized to derive a lower dimensional feature space from high-dimensional LIDAR waveform data. The resulting features were used to estimate mean canopy heights through multiple linear regression analysis. Experimental results obtained by the new approach were statistically comparable to estimates obtained using features extracted via traditional waveform analysis, and the proposed approach successfully discovered another meaningful lower dimensional feature space without manual interpretation.

COMPARISON OF EQUIVALENT LINEAR AND NON LINEAR METHODS ON GROUND RESPONSE ANALYSIS: CASE STUDY AT WEST BANGKA SITE. Within the framework of identifying NPP sites, site surveys are performed in West Bangka (WB), Bangka-Belitung Island Province. Ground response analysis of a potential site has been carried out using peak strain profiles and peak ground acceleration. The objective of this research is to compare Equivalent Linear (EQL) and Non Linear (NL) methods of ground response analysis on the selected NPP site (West Bangka) using DeepSoil software. Equivalent linear method is widely used because requires soil data in simple way and short time of computational process. On the other hand, non linear method is capable of representing the actual soil behaviour by considering non linear soil parameter. The results showed that EQL method has similar trends to NL method. At surface layer, the acceleration values for EQL and NL methods are resulted as 0.425g and 0.375g respectively. NL method is more reliable in capturing higher frequencies of spectral acceleration compared to EQL method.

A comparison was made of the linear least-squares method and a nonlinear least-square method of the Langmuir isotherm and pseudo second order kinetic model for the biosorption of methylene blue onto rice husk. The parameters obtained from the linear method and the nonlinear method were different. The error analysis was performed. Results showed that both linear and nonlinear method can be used to fit the experimental data. The nonlinear method may be a better way to obtain the desired parameters.

In this paper, we propose a manifold geometry based projective nonnegative matrix factorization linear dimensionality reduction method, called neighborhood preserving orthogonal projective nonnegative matrix factorization (NPOPNMF), for feature extraction of hyperspectral image. By adding constraints on projective nonnegative matrix factorization (PNMF) that each data point can be represented as a linear combination of its neighbors, NPOPNMF preserves local neighborhood geometrical structure of hyperspectral data in the reduced space, and overcomes the Euclidean limitation of PNMF. The metric structure of original high-dimensional hyperspectral data space is preserved due to the orthogonality of projection matrix. NPOPNMF can be performed in either supervised or unsupervised mode according to the construction of adjacency graph and it can improve the discriminant performance of PNMF. Theoretical analysis and experimental results on hyperspectral data sets demonstrate that the proposed method is an effective and promising method for hyperspectral image feature extraction.

A new nonlinear feature extraction method for face recognition

Because of the character of hyperspectral remote sensing data,it is necessary and urgent to develop hyperspectral data process algorithm.In this paper,some hyperspectral data process algorithms for feature selection and feature extraction was discussed,and its advantage disadvantage was analyzed.In particular,we studied derivative spectral algorithm and put forward quad-encoding algorithm as the improved the binary encoding algorithm.Using the algorithms this paper proposed we extract spectral absorption parameter.The experiments have demonstrated that quad-encoding algorithm has the better performance than binary encoding on hyperspectral data,and for derivative spectrum it is effective to indicate validate feature for objects when its rank is higher.

<p>Soil properties could be assessed with reflectance spectroscopy (soil spectroscopy, SS) in vis-NIR region (400-2500 nm) through absorption features found in soil spectra. A high spectral resolution (up to 1 nm) drives to high dimensional and multicollinear data. This issue is usually addressed prior to modelling with feature extraction methods such as Principal Component Analysis, or embedded methods such as Partial Least Squares Regression (PLSR). Feature Selection (FS) wrapper methods are promising dimensionality reduction approaches barely used in SS. The objective of this study was two-fold: i) evaluate the performance of FS wrapper methods built from Random Forest (RF) algorithm to predict soil organic matter (SOM), clay and carbonates using laboratory spectroscopy, ii) test the performance of FS methods for dimensionality reduction in SS. The reflectance of 100 soil samples from Sierra de las Nieves National Park (Spain), was measured under laboratory conditions using an ASD FieldSpec Pro JR. A spectral preprocessing method, Continumm Removal (CR), was applied to raw spectra. The RF wrapper considered two different feature searching approaches: Sequential Forward Selection (SFS) and Sequential Flotant Forward Selection (SFFS). The performance of RF with FS (RF-FS) was compared to that of Partial Least Squares Regression (PLSR) and RF (without FS). Models were evaluated with R-squared, root mean squared error (RMSE) and ratio of prediction to deviation (RPD).</p><p>RF-FS models outperformed PLSR and RF models for the three SAP. RF-FS best models had a RPD of 2.19 for SOM, 1.64 for carbonates and 1.52 for clay, whereas PLSR models had RPD values of 1.59, 1.22 and 1.3, and RF 1.38, 1.23 and 1.23 for SOM, carbonates, and clay, respectively. Therefore, FS was useful in obtaining models with improved accuracy by reducing redundant features and avoiding multicollinearity (Hughes effect). The application of FS wrapper methods reduced the number of features in the RF-FS models to less than 1% of the starting features. Features were selected across all spectra from SOM and clay, and around 900, 1900 and 2350 nm for carbonates. This research, thus, shows an alternative to different feature extraction approaches for modelling soil properties based on FS methods and machine learning.</p>

Recently, Hyperspectral remote sensing technology has been proved to be a valuable tool to get reliable information with details for identifying different objects on the earth surface with high spectral resolution. Due to atmospheric effects the valuable information may be lost from hyperspectral data. Hence it is necessary to remove these effects from hyperspectral data for reliable identification of the objects on the earth surface. The atmospheric correction is a very critical task of hyperspectral images. The present paper highlights the advantages of hyperspectral data, challenges over it as a pre-processing with solutions through QUAC and FLAASH algorithms. The hyperspectral data acquired for Aurangabad district were used to test these algorithms. The result indicates that the size of hyperspectral image can be reduced. The ENVI 5.1 software with IDL language is an efficient way to visualize and analysis the hyperspectral images. Implementation of atmospheric correction algorithms like QUAC and FLAASH is successfully carried out. The QUAC model gives accurate and reliable results without any ancillary information but requires only wavelength and radiometric calibration with less time than FLAASH.

This paper presents theoretical relationships among several generalized LDA algorithms and proposes computationally efficient approaches for them utilizing the relationships. Generalized LDA algorithms are extended nonlinearly by kernel methods resulting in nonlinear discriminant analysis. Performances and computational complexities of these linear and nonlinear discriminant analysis algorithms are compared.

The complex, nonlinear and non-stationary electroencephalogram (EEG) signals are very tedious to interpret visually and highly difficult to extract the significant features from them. The linear and nonlinear methods are effective in identifying the changes in EEG signals for the detection of depression. Linear methods do not exhibit the complex dynamical variations in the EEG signals. Hence, chaos theory and nonlinear dynamic methods are widely used in extracting the EEG signal features for computer-aided diagnosis (CAD) of depression. Hence, this article presents the recent efforts on CAD of depression using EEG signals with a focus on using nonlinear methods. Such a CAD system is simple to use and may be used by the clinicians as a tool to confirm their diagnosis. It should be of a particular value to enable the early detection of depression.

Traditional face recognition methods such as Principal Components Analysis(PCA), Independent Component Analysis(ICA) and Linear Discriminant Analysis(LDA) are linear discriminant methods, but in the real situation, a lot of problems can't be linear discriminated; therefore, researchers proposed face recognition method based on kernel techniques which can transform the nonlinear problem of inputting space into the linear problem of high dimensional space. In this paper, we propose a recognition method based on kernel function which combines kernel Fisher Discriminant Analysis(KFDA) with kernel Principle Components Analysis(KPCA) and use typical ORL(Olivetti Research Laboratory) face database as our experimental database. There are four key steps: constructing feature subspace, image projection, feature extraction and image recognition. We found that the recognition accuracy has been greatly improved by using nonlinear identification method and combined feature extraction methods. We use MATLAB software as the platform, and use the GUI to demonstrate the process of face recognition in order to achieving human-computer interaction and making the process and result more intuitive.

Face recognition as an important biometric identification method, with its friendly, natural, convenient advantages, has obtained more and more attention. This paper intends to research a face recognition system including face detection, feature extraction and face recognition, mainly through researching on related theory and the key technology of various preprocessing methods in face detection process, using KPCA method, focuses on the different recognition results in different preprocessing methods. In this paper, we choose YCbCr color space for skin segmentation and choose integral projection for face location. We use erosion and dilation of the opening and closing operation and illumination compensation method to preprocess face images, and then use the face recognition method based on kernel principal component analysis method for analysis and research, and the experiments were carried out using the typical face database. The algorithms experiment on MATLAB platform. Experimental results show that integration of the kernel method based on PCA algorithm under certain conditions make the extracted features represent the original image information better for using nonlinear feature extraction method, which can obtain higher recognition rate. In the image preprocessing stage, we found that images under various operations may appear different results, so as to obtain different recognition rate in recognition stage. At the same time, in the process of the kernel principal component analysis, the value of the power of the polynomial function can affect the recognition result.