Rapid and non-destructive measurement of biofuel pellet quality indices based on two-dimensional near infrared spectroscopic imaging

Rapid measurements for characterizing biomass pellets and monitoring the production process are needed due to increasing use of biofuel pellets in the context of renewable energy. The objective of this study was to assess the accuracy and reliability of hyperspectral imaging combined with chemometrics to predict biofuel pellet properties for automatic online application. First, a successive projections algorithm (SPA) was applied to hyperspectral image analysis to select the important variables. Subsequently, prediction models established using partial least-squares regression (PLSR) and a least-squares support vector machine (LSSVM) based on whole wavelengths and important wavelengths were compared. The optimized prediction models constructed by SPA-LSSVM showed excellent performance for measuring moisture content, ash content, volatile matter, and calorific value, with the coefficients of determination being 0.94, 0.92, 0.94, and 0.90 respectively. Finally, all the quality indices were quantitatively visualized on prediction maps by transferring each optimal model to each pixel in the hyperspectral images. The results show that calibration models for biofuel pellets quality indices are successful developed, which would enable pellet producers to improve the operation of the pelletizing process for high-throughput applications.