Machine Learning-Based Prediction of Selected Parameters of Commercial Biomass Pellets Using Line Scan Near Infrared-Hyperspectral Image

Biomass pellets are required as a source of energy because their abundant and high energy. The rapid measurement is used to control the biomass quality during production process. objective this work was use near infrared (NIR) hyperspectral images for predicting properties, i.e., fuel ratio (FR), volatile matter (VM), fixed carbon (FC), ash content (A), commercial pellets. Models were developed using either full spectra or different spatial wavelengths, interval successive projections algorithm (iSPA) genetic (iGA), wavelengths spectral preprocessing techniques. Their performances then compared. optimal model FR could be created with second derivative (D2) iSPA-100 while VM, FC, A predicted standard normal variate (SNV) wavelengths. models FR, provided R2 values 0.75, 0.81, 0.82, 0.87, respectively. Finally, prediction pellets’ properties under color distribution mapping able track pellet monitor operation thermal conversion process can intuitively applications screening.