Bananas moisture content prediction system using Visual-NIR imaging

Commonly, the fruit moisture content could be measured by comparing the mass decrement of an object through the oven drying method. Bananas moisture content prediction system was introduced using the Visual-NIR imaging technique. The hardware of the proposed system consists of a set of the workbench, controllable slider, two halogen light sources and a hyperspectral camera that connected to PC via Camera Link. The software of system consists of reflectance image profile measurement, feature extraction on spectral and spatial data, and moisture prediction model. The reflectance image profile was extracted from the banana surface based on current image, white and dark image reference. The feature sets were computed using a principal component analysis (PCA) and partial least square regression (PLSR) on the full wavelength range of HSI spectra. The purpose of using two regression methods in this research is for comparing the result of moisture content prediction. The proposed system was evaluated using 45 Raja bananas (Musa textilla) samples which consist of 15 samples for each maturity stage. The prediction error between predicted and measured data with PCR is 0.58 % and produce correlation coefficient R² of 0.79. The PLSR model of banana content prediction system has RMSE 0.25% and R² 0.96. The results show that the proposed system can predict the banana moisture content and suitable in an industrial sorting system for banana fruit quality.