Achieving robustness to temperature change of a NIR model for apple soluble solids content

Abstract

The temperature difference of fruit itself will affect its near infrared (NIR) spectrum and the accuracy of its soluble solids content (SSC) prediction model. To eliminate the influence of apple temperature difference on SSC model, the diffuse transmission dynamic online detection device was used to collect the spectral data of apples at different temperatures, and four methods were used to establish partial least squares (PLS) correction models: global correction, orthogonal signal processing (OSC), generalized least squares weighting (GLSW) and external parameter orthogonal (EPO). The results show that the temperature has a strong influence on the diffuse transmission spectrum of apples. The 20℃ model can get a satisfactory prediction result when the temperature is constant, and there will be great errors when detecting samples at other temperatures. The effect of temperature must be corrected to establish a more general model. These methods all improve the accuracy of the model, with the EPO method giving the best results, prediction set correlation coefficient (Rp) is 0.947, Root Mean Square Error of Prediction (RMSEP) is 0.489 %Brix, and the prediction bias (Pred Bias) is 0.009 %Brix. The research results are of great significance to the practical application of SSC prediction of fruits in sorting workshops or orchards.