Prediction of the anthocyanin content of Lonicera edulis based on fruit color characteristics

Abstract: Anthocyanin is one of the most landmark components in Lonicera japonica Thunb. The content of anthocyanin can be utilized to determine the timely harvest time and the quality of fresh fruit. The chemical methods are commonly used to detect the anthocyanin content in berries under laboratory conditions, including spectrophotometry, pH difference, and chromatography. The standard determination of anthocyanins in fruits and vegetables can present high accuracy and wide application. But, online and real-time detection is limited to the long time consuming and strong specialization. It is feasible to predict and evaluate the internal quality of agricultural products, according to their appearance and color characteristics. Taking the Lonicera edulis fruit as the research object, this study aims to determine the relationship between the appearance color indexes and the anthocyanins content under different growth periods. Gaussian processes regression (GPR) model was established to characterize the anthocyanin content as the function of characteristic color indexes of Lonicera edulis fruits. The nondestructive detection of anthocyanin content was also realized in Lonicera edulis fruit. The results indicate that the anthocyanin content of fruit increased rapidly from the fourth stage of Lonicera edulis growth, and the second stage of half color transition. The highest level was achieved at the seventh stage of maturity, during which the fruit color changed from the red to the purple black. During the growth and ripening of Lonicera edulis fruit, 21 color characteristic parameters were formed by the single component of RGB tri-color and HSV color saturation of the fruit and the combined component obtained by arithmetic operation. These parameters were significantly correlated with anthocyanin content (P<0.01); Based on the normalized difference of fruit green and blue primary colors, the optimal Gaussian model of anthocyanin content was established with high accuracy (R2c=0.993) and low error (CRMSE=4.752 mg/100g). The typical prediction models of anthocyanin content in Lonicera edulis included multiple linear regression, neural network, Gaussian function, and support vector machine. A comparison was made for the main evaluation indicators of the typical modeling, from the aspects of the fitting degree, mean square error, the number of required factors, and the complexity of the model. The high prediction accuracy, low deviation, and simple structure were achieved in the Gaussian function model with the characteristic color value of Lonicera edulis to predict the anthocyanin content, indicating superior to the other three prediction models. In particular, the prediction of anthocyanin content with one characteristic parameter g-b (X21) reduced the illumination interference of environmental factors, such as reflection, indicating the practical significance. Multivariate linear regression model and support vector machine model presented a lower fitting degree, more required factors, and more complicated color information processing, but these two models also had higher practicability. In the artificial neural network (ANN) model, all 21 characteristic parameters were required to participate with the low fitting degree, indicating the relatively low applicability at present. The optimal Gaussian prediction model performed fast with less amount of data, suitable for the prediction of the anthocyanin content of Lonicera edulis fruit. As such, the real-time evaluation of interesting quality indicators can be realized after the non-destructive determination of anthocyanin content during the growth period of Lonicera edulis. The finding can also provide guidance for the determination of rapid, non-destructive, and high-accuracy detection and equipment design of anthocyanin content in Lonicera edulis during growth and harvest.