Subpixel Classification on the Hyperspectral Images for Accuracy Improvement of Classification Results

Classification is a common and most important technique for information extraction, from remotely sensed images. In traditional classification methods, each pixel is assigned to a single class by presuming all pixels within the image are pure. Therefore, this is the main problem and limitation of traditional image classification procedures in classification of mixed pixel. Mixed pixel classification is a process which tries to extract the proportions of the pure components of each mixed pixel. This approach is known as spectral unmixing. Spectral unmixing is a method which allows the user to determine information on a subpixel level and to study decomposition of mixed pixels. Hyperspectral images have the high spectral resolution rather than to multispectral images. By development of remote sensing technology, the new sensors with hyperspectral capabilities in RS science will be replaced to multispectral imaging. A big advantage of hyperspectral images comparison to that of multispectral images is a continuous spectrum for each image cell that can be derived from image spectral measurement. Therefore, in this research these images have been used in classification process. In this paper, pixel-based classification methods including the spectral angle mapper, maximum likelihood classification and subpixel classification method (linear spectral unmixing) were implemented on the AVIRIS hyperspectral images. Then, a comparison between pixel-based and subpixel based classification algorithms was carried out. Also, this paper investigates the capabilities and advantages of spectral linear unmixing method. The spectral unmixing method that implemented here is an effective technique for classifying a hyperspectral image giving the classification accuracy about 89%, while in pixel based classification methods implementing on hyperspectral images, the maximum accuracy is about 74% for spectral angle mapper and 81 % for maximum likelihood classification methods. The results of classification when applying on the original images are not good because some of the hyperspectral image bands are subject to absorption and they contain only little signal but more noisy. Therefore, preparation of data is necessary at the beginning of process. For this purpose, by applying the Minimum Noise Fraction (MNF) transformation on the hyperspectral images, the correlation and noises from bands can be removed and these bands can be sorted in respect of their variance. In bands with a high variance, the features can be distinguished from each other in a better mode, therefore classification accuracy is increased. Also, applying the MNF transformation on the hyperspectral images increase the individual classes accuracy of pixel based classification methods as well as unmixing method about 20 percent and 9 percent respectively.

Keywords:

Accuracy Improvement , Hyperspectral Images , Minimum Noise Fraction Transformation , Mixed pixels , Spectral Unmixing , Sub-pixel Classification