Classification of Hyper Spectral Image Various Plant Classes via Coding Method in the Reflectance and its Derivatives

Plants paly a very important role in creating and maintaining the biological balance which is vital for the life of each living creature including humans. Due to the great importance of vegetation cover in terms of habitat, energy production and other important characteristics on the planet, the recognition and monitoring of various plant species has always been a concern for ecologists and decision makers from the economic points of view. This will be possible on a large scale with remote sensing technology and the use of satellite images containing vegetation and their classification. Different techniques of classification are deployed where in some of them, use have been made of reflectance curves and their derivatives. In some other methods, some coding system applied on reflectance curves and their derivatives are used as a fast method.  In this work, a method named CBOSE is presented in which, a coding approach on reflectance and its derivatives is applied. The CBOSE method is coding based on extreme points of the reflectance spectrum and combines from one to several bits to distinguish between plant species with relatively high spectral similarity. This coding method, after necessary pre-processing such as water vapor correction and continuum removal analysis, on AVIRIS hyperspectral images of Indian pine region containing various species such as wheat, barley, alfalfa, grass, tree, soybean and corn were also applied in three stages of germination, medium growth and full growth. Then, the features with the highest separability between the classes were extracted and the classification was done on the properties derived from the codes by selecting the training samples. The classification output of CBOSE was compared with the result of classification by classifiers Support Vector Machine (SVM), Maximum likelihood (ML), Spectral Angle Mesure (SAM), and Hamming similarity criteria and with those of field data. Also the methodology of CBOSE was evaluated and compared with those of coding methods such as Spectral analysis manager (SPAM), Spectral feature-based binary coding (SFBC), Spectral derivative feature coding (SDFC), and Spectral feature probabilistic coding) SFPC). The results show that the CBOSE methods on the average performs respectively 20, 16, 11 and 7 percent better compared to the afore-mentioned methods. Also, in order to evaluate the effects of using derivatives in the coding process, all aforementioned procedures were repeated without using derivatives in the coding processes. It showed that on the average, deployment of reflectance derivative would 8% enhances the accuracy in classification.