Nondestructive prediction model of internal hardness attribute of fig fruit using NIR spectroscopy and RF

Abstract

Hardness is one of the most important quality characteristics, which has an important influence on the processing and product quality of figs. A rapid non-destructive detection method for the hardness of figs was proposed based on visible/near infrared (VIS/NIR) spectroscopy technology. This study attempts to optimize the construction of a fig hardness model and predict the accuracy of thereof. An NIR spectrometer was used to collect the diffuse reflectance spectrum data in the wavelength range of 950–1700 nm, while the hardness index was measured using texture analyzer. Random forest (RF) and partial least square (PLS) methods were used to model the spectral data and hardness, respectively, and a better algorithm for the model construction was obtained. The RF model performed better in the characteristic band (1150.83–1232.43 nm), with correlation coefficient (R2), root mean square error of calibration (RMSEC), and root mean square error of prediction (RMSEP) of 0.76, 67.61, and 83.94 respectively. The PLS model worked well at the full band (R2 = 0.77, RMSEC = 59.20, RMSEP = 91.84). However, the prediction time of the PLS was slightly shorter than that of RF model (0.0004 s < 0.0098 s). The results show that it is feasible to detect the hardness of figs without destroying them by using VIS/NIR diffuse reflectance spectroscopy combined with sample set partitioning based on joint x–y distances (SPXY), RF, and PLS algorithms. This study provides new technical means for fig products enterprises to determine the hardness of figs in the early stages of production rapidly and evaluate the processing quality of fig products, which has a high practical application potential.