Determination Lactone Composition in Andrographis paniculata(burm.f.) Wall, ex Nees using hyperspectral imaging

Nowadays, hyperspectral images (HSI) have been widely used in chemometrics field. Partial least square regression (PLSR) and principal component regression model (PCR) were dimensional reduction applied to solve the large p small n problem that is commonly found in HSIs. In this study, the main objective is to establish a prediction model in order to estimate the lactone content in Andrographis paniculata. Spectral reflectance of HSIs within wavelengths between 355 – 1505 nm. was extracted from pixels of plants that are separated from the background by the normalized difference vegetation index (NDVI). The pre-processing method including the first and second derivatives of the Savitzky-Golay filter was used to eliminate the noise that comes along with raw spectral data. To select optimal wavelength regression coefficient (RC), forward variable selection (FVS), and backward variable elimination (BVE) were employed in this study. After feature selection, more than 88% of spectral variables were eliminated. Models that were performed in this study indicated the model that pre-processing spectral with the second derivative and variable selection by BVE yields better performance than the other methods with Rp of 0.971, RMSEP of 1.050, RCV of 0.995, and RMSECV of 0.606. The results show that using hyperspectral images combinate with the PLSR model can perform a reliable and accurate prediction model for determining lactone content in A. paniculata.