Bin Pan

R-VCANet: A New Deep-Learning-Based Hyperspectral Image Classification Method

Deep-learning-based methods have displayed promising performance for hyperspectral image (HSI) classification, due to their capacity of extracting deep features from HSI. However, these methods usually require a large number of training samples. It is quite difficult for deep-learning model to provide representative feature expression for HSI data when the number of samples are limited. In this paper, a novel simplified deep-learning model, rolling guidance filter (RGF) and vertex component analysis network (R-VCANet), is proposed, which achieves higher accuracy when the number of training samples is not abundant. In R-VCANet, the inherent properties of HSI data, spatial information and spectral characteristics, are utilized to construct the network. And by this means the obtained model could generate more powerful feature expression with less samples. First, spectral and spatial information are combined via the RGF, which could explore the contextual structure features and remove small details from HSI. More importantly, we have designed a new network called vertex component analysis network for deep features extraction from the smoothed HSI. Experiments on three popular datasets indicate that the proposed R-VCANet based method reveals better performance than some state-of-the-art methods, especially when the training samples available are not abundant.

Hyperspectral Image Classification Based on Nonlinear Spectral–Spatial Network

Recently, for the task of hyperspectral image classification, deep-learning-based methods have revealed promising performance. However, the complex network structure and the time-consuming training process have restricted their applications. In this letter, we construct a much simpler network, i.e., the nonlinear spectral-spatial network (NSSNet), for hyperspectral image classification. NSSNet is developed from the basic structure of a principal component analysis network. Nonlinear information is included in NSSNet, to generate a more discriminative feature expression. Moreover, spectral and spatial features are combined to further improve the classification accuracy. Experimental results indicate that our method achieves better performance than state-of-the-art deep-learning-based methods.

Hierarchical Guidance Filtering-Based Ensemble Classification for Hyperspectral Images

Joint spectral and spatial information should be fully exploited in order to achieve accurate classification results for hyperspectral images. In this paper, we propose an ensemble framework, which combines spectral and spatial information in different scales. The motivation of the proposed method derives from the basic idea: by integrating many individual learners, ensemble learning can achieve better generalization ability than a single learner. In the proposed work, the individual learners are obtained by joint spectral-spatial features generated from different scales. Specially, we develop two techniques to construct the ensemble model, namely, hierarchical guidance filtering (HGF) and matrix of spectral angle distance (mSAD). HGF and mSAD are combined via a weighted ensemble strategy. HGF is a hierarchical edge-preserving filtering operation, which could produce diverse sample sets. Meanwhile, in each hierarchy, a different spatial contextual information is extracted. With the increase of hierarchy, the pixels spectra tend smooth, while the spatial features are enhanced. Based on the outputs of HGF, a series of classifiers can be obtained. Subsequently, we define a low-rank matrix, mSAD, to measure the diversity among training samples in each hierarchy. Finally, an ensemble strategy is proposed using the obtained individual classifiers and mSAD. We term the proposed method as HiFi-We. Experiments are conducted on two popular data sets, Indian Pines and Pavia University, as well as a challenging hyperspectral data set used in 2014 Data Fusion Contest (GRSS_DFC_2014). An effectiveness analysis about the ensemble strategy is also displayed.

A Novel Spectral-Unmixing-Based Green Algae Area Estimation Method for GOCI Data

Geostationary Ocean Color Imager (GOCI) data have been widely used in the detection and area estimation of green algae blooms. However, due to the low spatial resolution of GOCI data, pixels in an image are usually “mixed,” which means that the region a pixel covers may include many different materials. Traditional index-based methods can detect whether there are green algal blooms in each pixel, whereas it is still challenging to determine the proportion that green algae blooms occupy in a pixel. In this paper, we propose a novel subpixel-level area estimation method for green algae blooms based on spectral unmixing, which can not only detect the existence of green algae but also determine their proportion in each pixel. A fast endmember extraction method is proposed to automatically calculate the endmember spectral matrix, and the abundance map of green algae which could be regarded as the area estimation is obtained by nonnegatively constrained least squares. This new fast endmembers extraction technique outperforms the classical N-FINDR method by applying two models: candidates location and distance-based vertices determination. In the first model, we propose a medium-distance-based candidates location strategy, which could reduce the searching space during vertices selection. In the second model, we replace the simplex volume measure with a more simple distance measure, thus complex matrix determinant calculation is avoided. We have theoretically proven the equivalency of volume and distance measure. Experiments on GOCI data and synthetic data demonstrate the superiority of the proposed method compared with some state-of-art approaches.

A New Unsupervised Hyperspectral Band Selection Method Based on Multiobjective Optimization

Unsupervised band selection methods usually assume specific optimization objectives, which may include band or spatial relationship. However, since one objective could only represent parts of hyperspectral characteristics, it is difficult to determine which objective is the most appropriate. In this letter, we propose a new multiobjective optimization-based band selection method, which is able to simultaneously optimize several objectives. The hyperspectral band selection is transformed into a combinational optimization problem, where each band is represented by a binary code. More importantly, to overcome the problem of unique solution selection in traditional multiobjective methods, we develop a new incorporated rank-based solution set concentration approach in the process of Tchebycheff decomposition. The performance of our method is evaluated under the application of hyperspectral imagery classification. Three recently proposed band selection methods are compared.

Hyperspectral Image Classification Based on Deep Forest and Spectral-Spatial Cooperative Feature

Recently, deep-learning-based methods have displayed promising performance for hyperspectral image (HSI) classification. However, these methods usually require a large number of training samples, and the complex structure and time-consuming problem have restricted their applications. Deep forest, a decision tree ensemble approach with performance highly competitive to deep neural networks. Deep forest can work well and efficiently even when there are only small-scale training data. In this paper, a novel simplified deep framework is proposed, which achieves higher accuracy when the number of training samples is small. We propose the framework which employs local binary patterns (LBPS) and gabor filter to extract local-global image features. The extracted feature along with original spectral features will be stacked, which can achieve concatenation of multiple features. Finally, deep forest will extract deeper features and use strategy of layer-by-layer voting for HSI classification.

Hashing Based Hierarchical Feature Representation for Hyperspectral Imagery Classification

Integrating spectral and spatial information is proved effective in improving the accuracy of hyperspectral imagery classification. In recent studies, two kinds of approaches are widely investigated: (1) developing a multiple feature fusion (MFF) strategy; and (2) designing a powerful spectral-spatial feature extraction (FE) algorithm. In this paper, we combine the advantages of MFF and FE, and propose an ensemble based feature representation method for hyperspectral imagery classification, which aims at generating a hierarchical feature representation for the original hyperspectral data. The proposed method is composed of three cascaded layers: firstly, multiple features, including local, global and spectral, are extracted from the hyperspectral data. Next, a new hashing based feature representation method is proposed and conducted on the features obtained in the first layer. Finally, a simple but efficient extreme learning machine classifier is employed to get the classification results. To some extent, the proposed method is a combination of MFF and FE: instead of feature fusion or single feature extraction, we use an ensemble strategy to provide a hierarchical feature representation for the hyperspectral data. In the experiments, we select two popular and one challenging hyperspectral data sets for evaluation, and six recently proposed methods are compared. The proposed method achieves respectively 89.55%, 99.36% and 77.90% overall accuracies in the three data sets with 20 training samples per class. The results prove that the performance of the proposed method is superior to some MFF and FE based ones.

Longwave infrared hyperspectral image classification via an ensemble method

ABSTRACT Longwave infrared hyperspectral images (LWIR-HSIs) classification is challenging, due to the poor imaging quality and low signal-to-noise ratio. A popular viewpoint is that abundant spatial contextual information can significantly improve the classification accuracies. However, it is quite difficult to determine what degree of spatial information is the most useful. In this article, we develop a novel ensemble-based classification method, which is able to fully leverage joint spectral-spatial features in different degrees. The proposed method contains three primary steps. First, a powerful edge-preserving filtering (EPF) approach, rolling guidance filtering (RGF), is utilized to generate several groups of diverse samples as well as enhance the quality of the LWIR-HSI data. Each group corresponds to a certain degree of spatial information. Subsequently, a series of individual classifiers are learned based on all groups of training samples, and each classifier could provide a single classification result for all test samples. Finally, we propose a new ensemble strategy, multi-classifier -statistic (MKS), to evaluate the contributions of individual learners (ILs). The final classification results are obtained based on the outputs of RGF and MKS. Experiments on a challenging LWIR-HSI data set verify the effectiveness of the proposed method, compared with some state-of-the-art HSI classification methods.