Unsupervised classification method for hyperspectral image combining PCA and Gaussian mixture model

Wu, Hao; Kuang, Gangyao; Wang, Yu

doi:10.1117/12.539872

Cited by 7 publications

(3 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our point of view, dimension reduction is a paradox: hyperspectral sensors are developed to collect rich information of the surface for increasing the discrimination of materials on the surface, while dimension reduction works on abbreviating the information in real classification. Many classification algorithms like unsupervised classification methods [12] which don't require training samples but the clusters produced rarely maps correctly to the correct classes. In supervised methods category we have artificial neural networks [5] but a lot of training samples are required for these and does not work well for high dimensional data.…”

Section: Introductionmentioning

confidence: 99%

A l1-minimization Based Approach for Hyperspectral Data Classiﬁcation

Qazi

Shi

Tao

et al. 2012

KEM

View full text Add to dashboard Cite

Most of classiﬁcation methods require model learning procedure or optimal parameters selection using large number of training samples. In this paper, we propose a novel classiﬁcation approach using l1-minimization based sparse representation which does not need any learning procedure or parameters selection. The proposed approach is based on l1minimization because l0-minimization is generally NP-hard and is not a convex optimization problem. Sparse based solutions have been proposed in other areas like signal processing but to the best of our knowledge, this is the ﬁrst time that sparse based classiﬁcation is being proposed for the classiﬁcation of hyperspectral data. l1-minimization based sparse representation is calculated for each of the test samples using few training samples directly. In remote sensing, usually it is extremely difﬁcult and expensive to identify and label the samples due to which we often lack sufﬁcient training samples for classiﬁcation. We tested the proposed approach for difﬁcult classiﬁcation problem i.e high dimensional spaces and few training samples. We also analyzed the cases in which the sparse based classiﬁcation may and may not work. Our extensive experiments on real hyperspectral dataset (AVIRIS 1992 Indiana’s Indian Pines image), prove that the proposed technique offers more classiﬁcation accuracy and more efﬁcient than state-of-the-art SVM and semi-supervised methods as it does not need any model selection.

show abstract

Section: Introductionmentioning

confidence: 99%

A l1-minimization Based Approach for Hyperspectral Data Classiﬁcation

Qazi

Shi

Tao

et al. 2012

KEM

View full text Add to dashboard Cite

show abstract

“…In [13], an unsupervised method based on Gaussian mixture model is proposed. In [17], an unsupervised classification method is proposed which is based on the combination of PCA and Gaussian mixture models. These clusters to classes mapping rarely work correctly in general.…”

Section: Introductionmentioning

confidence: 99%

Neural network based adaboosting approach for hyperspectral data classification

Haq

Tao

2011

Proceedings of 2011 International Conference on Computer Science and Network Technology

View full text Add to dashboard Cite

In this paper, we propose a novel approach for hyperspectral data classification using adaboosting of artificial neural networks based weak classifiers. The adaboost algorithm employs an iterative approach which combines weak classifiers to approximate a Bayes classifier. It requires performance of each weak classifier to be a little better than random guessing. In the approach, we use several neural network based weak classifiers to make one strong classifier and each weak classifier contains only one hidden layer. As the weak classifiers are simple, therefore, these does not require a lot of time for training and, therefore, are time efficient. Back propagation algorithm is used for the learning purposes. In the approach, weights are assigned to each training sample and those are boosted in subsequent iterations for the misclassified training samples, therefore, providing more robust approach to classify difficult samples. A confidence score is assigned to each weak classifier. Final classifier is achieved by linear combination of all the weak classifiers. Entropy gain clustered PCA is used for the dimension reduction of the data. We performed experiments on real hyperspectral dataset of AVIRIS Indian Pines and the comparisons with neural network and ML based approaches prove the efficiency of the proposed approach.

show abstract

“…Experimental studies have shown that the performance of wavelets in classification is very good for vegetation detection with Airborne Visible/Infrared Imaging Spectrometer data [6] and hyperspectral data collected at ground level [7]. The use of unsupervised classification methods like PCA [8] or hierarchical clustering [9] is limited when processing large 0196-2892/$25.00 © 2007 IEEE TABLE I BASIC SPECTRAL CHARACTERISTICS OF THE OMEGA INSTRUMENT ONBOARD MEX datasets because such methods are computer intensive. Furthermore, they do not take advantage of the a priori knowledge we may have concerning planetary surfaces and atmospheres.…”

Section: Introductionmentioning

confidence: 99%

WAVANGLET: An Efficient Supervised Classifier for Hyperspectral Images

Schmidt

Douté

Schmitt

2007

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Abstract-The new generation of imaging spectrometers onboard planetary missions usually produce hundreds to thousands of images a year, each made up of a thousand to a million spectra with typically several hundred wavelengths. Such huge datasets must be analyzed by efficient yet accurate algorithms. A supervised automatic classification method (hereafter called "wavanglet") is proposed to identify spectral features and classify images in spectrally homogeneous units. It uses four steps: 1) selection of a library composed of reference spectra; 2) application of a Daubechies wavelet transform to referenced spectra and determination of the wavelet subspace that best separates all referenced spectra; and 3) in this selected subspace, determination of the best threshold on the spectral angle to produce detection masks. This application is focused on the Martian polar regions that present three main types of terrains: H 2 O ice, CO 2 ice, and dust. The wavanglet method is implemented to detect these major compounds on near-infrared hyperspectral images acquired by the OMEGA instrument onboard the Mars Express spacecraft. With an overall accuracy of 89%, wavanglet outperforms two generic methods: band ratio (57% accuracy) and spectral feature fitting (83% accuracy). The quantitative detection limits of wavanglet are also evaluated in terms of abundance for H 2 O and CO 2 ices in order to improve the interpretation of the masks.

show abstract

Unsupervised classification method for hyperspectral image combining PCA and Gaussian mixture model

Cited by 7 publications

References 5 publications

A l<sup>1</sup><i>-minimization </i>Based Approach for Hyperspectral Data Classiﬁcation

A l<sup>1</sup><i>-minimization </i>Based Approach for Hyperspectral Data Classiﬁcation

Neural network based adaboosting approach for hyperspectral data classification

WAVANGLET: An Efficient Supervised Classifier for Hyperspectral Images

Contact Info

Product

Resources

About