Hyperspectral imaging acquisition and calibration

For each biofuel pellet sample, about 10 g was selected for hyperspectral image acquisition. The apparatus of NIR hyperspectral imaging for acquiring the images was as described by Feng et al. [16]. Figure 1 illustrates the typical push-broom system. The instruments included an ImSpector N17E imaging spectrograph (Specim, Oulu, Finland) using the wavelengths 900–1700 nm, two 150-W halogen lamps (Fiber-Lite DC9500 Illuminator; Dolan Jenner Industries Inc., USA), a CCD camera (Xeva 992; Xenices Infrared Solutions, Leuven, Belgium) and a C-mount imaging lens (OLES22; Specim). Three instrument parameters had to be reset before obtaining a clear and non-deformed image. In the present study, biofuel pellets were placed on the conveyer belt and moved at the speed of 3.1 mm/s. The exposure time of the camera was 0.03 s. The vertical distance between lens and samples was 31.2 cm. Each raw hyperspectral image of a biofuel pellet contained 256 contiguous wavelength bands and 256 congruent sub-images.

Configuration of the hyperspectral imaging system and flowchart of hyperspectral image segmentation

The raw hyperspectral images (I_raw) were calculated using a white (I_white) and a dark reference image (I_dark) to correct the reflectance spectrum from the illumination and device response. The calibrated image (I_cal) was calculated using Eq. (1):

where I_dark was acquired by turning light sources off and covering the camera with reflectance close to 0%, and I_white was obtained using white Teflon with ~ 99% reflectance under the same condition.