Model Research On Efficient Detection Of Nutritional Components In Blueberry Pomace Using Visible/Near-infrared Spectroscopy

Blueberry pomace is rich in anthocyanins,polyphenols,dietary fiber,and other nutrients,making it an important source of natural antioxidants and pigments with significant application value.As a valuable raw material,the content of anthocyanins and total phenolics of blueberry pomace are important indicators of its utilization value.However,traditional chemical detection methods are time-consuming and complex,making it difficult to make rapid judgments on the utilization value of blueberry pomace.With the development of visible/nearinfrared spectroscopy technology,spectroscopic methods for detecting nutritional components have been widely used in the food industry.In this paper,the optimal detection model of anthocyanins and total phenolics in blueberry pomace based on visible/near-infrared spectroscopy was established,enabling efficient detection and providing a basis for accurate and rapid judgment of the utilization value of blueberry pomace.This can achieve the full utilization of blueberry pomace,increase its added value and economic benefits.This paper focuses on rabbit-eye blueberry pomace and uses a portable spectrometer to collect visible/near-infrared spectroscopy data of blueberry pomace experimental samples.At the same time,physochemical experiments are used to determine the anthocyanin and total phenolic content of blueberry pomace samples to obtain 150 sets of experimental data.The collected spectral data are preprocessed using convolution smooth,multi-scatter correction,standard normal transformation,first-order derivative,and trend removal correction methods,and feature selection is performed using competitive adaptive reweighted sampling and Pearson correlation analysis.The 150 experimental samples are divided into 113 calibration set samples and 37 test set samples using the leave-one-out strategy.By inputting the source data,preprocessed data,and feature-selected data into partial least squares regression,extreme learning machine,least squares support vector regression,and stacked supervised autoencoder,four detection models of anthocyanin and total phenolic content in blueberry pomace are constructed.Using the correlation coefficient(Rp),root mean square error of prediction(RMSEP),and mean absolute error(MAE)as evaluation indicators,the comparative detection results of the four models show that in detecting the anthocyanin content of blueberry pomace,the stacked supervised autoencoder model achieves the highest accuracy.The built model obtained an Rp of 0.9042,RMSEP of 0.3835,MAE of 0.3039,and an average relative error of 6.37% on the test set.However,in detecting the total phenolic content of blueberry pomace,the extreme learning machine model achieves the highest accuracy,with an Rp of 0.8792,RMSEP of0.4712,MAE of 0.3371,and an average relative error of 6.83%.The study shows that using visible/near-infrared spectroscopy technology in combination with stacked supervised autoencoder and extreme learning machine is an efficient approach for the detection of anthocyanin and total phenolic content in blueberry pomace.