Sweetness quantitative spectral measurement of apple based on visible light spectroscopy combined with BP neural network

Abstract

In this study, the visible light spectroscopy was used to achieve the sweetness quantitative measurement of apple. In the experiments, the absorption spectra of apple samples in total of 100 groups were obtained in the waveband from 400-800nm with interval of 5nm by using the visible light spectroscopy. At the same time, the real sweetness values of all apples were measured by using a commercial fruit sugar meter. To achieve the sweetness quantitative spectral measurement, the back propagation (BP) neural network was used to supervised train the absorption spectral for 80 groups of training samples, and 20 groups of apples were utilized as the test samples. The effects of neuron numbers in the hidden layer, learning rate factor and the training times on the root-mean-square error (RMSE) of sweetness were investigated. Under the optimal parameters of BP neural network, the RMSE of sweetness for the test apple samples can reach 0.12218%, which is superior to that of the commercial fruit sugar meter (0.2%). Compared with the correlation coefficients for the training samples and test samples based on the partial least square (PLS) algorithm, it can be demonstrated that the visible light spectroscopy combined with BP neural network has the potential superiority and application value in the sweetness quantitative spectral measurement of fruit.