Sweetness classification of apple based on VIS spectroscopy combined with particle swarm optimized BP neural network

In this study, 100 groups of apples with different sweetness were measured in transmission mode using visible light spectroscopy (VIS). The absorption spectra of all samples were obtained in the wavelength range of 400-800 nm with a step of 5 nm. To classify and identify the sweetness of apples, a qualitative classification model of apple absorption spectra and sweetness was constructed using BP neural network. The sweetness of all apples was classified into three different classes and labeled with Arabic numbers from one to three. In the experiment, 80 groups of apples were randomly selected as training samples and 20 groups of apples as test samples. Through the test, the sweetness classification accuracy of the test samples based on BP neural network reached 75%. To further improve the classification accuracy of sweetness, a Particle Swarm Optimization (PSO) algorithm was used to optimize the parameters of the BP neural network. With the optimal values of BP-PSO model parameters, the sweetness classification accuracy reached 90% for 20 sets of test samples. Finally, traditional classification models of extreme learning machine (ELM), competitive neural network (CNN) and self-organizing mapping neural network (SOMNN) were established to compare the classification accuracy of different algorithms, and the accuracy of 50%, 35% and 65% was achieved using ELM, CNN and SOMNN models, respectively. The results show that the classification using BP-PSO model has higher classification accuracy. Therefore, the BP-PSO model can be applied to the quality identification and classification of apples based on VIS technique.