Three-Dimensional Spatial-Spectral Filtering Based Feature Extraction for Hyperspectral Image Classification

• Published in: Advances in electrical and computer engineering Vol. 17; no. 2; pp. 95 - 102
• Main Authors: AKYUREK, H. A., KOCER, B.
• Format: Journal Article
• Language: English
• Published: Suceava Stefan cel Mare University of Suceava 01-05-2017
• Subjects: Accuracy; adaptive algorithms; Algorithms; Classification; Feature extraction; Filtration; gaussian noise; Hyperspectral imaging; Image classification; Methods; Neighborhoods; Noise; Pixels; Remote sensing; Researchers; Spatial data; Spectra; Spectral resolution; Support vector machines
• ISSN: 1582-7445; 1844-7600
• DOI: 10.4316/AECE.2017.02013

Hyperspectral pixels which have high spectral resolution are used to predict decomposition of material types on area of obtained image. Due to its multidimensional form, hyperspectral image classification is a challenging task. Hyperspectral images are also affected by radiometric noise. In order to improve the classification accuracy, many researchers are focusing on the improvement of filtering, feature extraction and classification methods. In the context of hyperspectral image classification, spatial information is as important as spectral information. In this study, a three-dimensional spatialspectral filtering based feature extraction method is presented. It consists of three main steps. The first is a pre-processing step, which include spatial-spectral information Altering in three-dimensional space. The second comprises extract functional features of filtered data. The last one is combining extracted features by serial feature fusion strategy and using to classify hyperspectral image pixels. Experiments were conducted on two popular public hyperspectral remote sensing image, 1%, 5%, 10% and 15% of samples of each classes used as training set, the remaining is used as test set. The proposed method compared with well-known methods. Experimental results show that the proposed method achieved outstanding performance than compared methods in hyperspectral image classification task.