Nondestructive prediction of fruit detachment force for investigating postharvest grape abscission

The distinct flavor and beneficial nutritional qualities of table grapes make them a top choice among customers. However, due to natural senescence, environmental stress, and excessive SO2 preservatives, grapes are prone to abscission after harvest, which increases harvest losses, lowers fruit quality, and reduces economic value. A primary cause of grape abscission is a decrease in fruit detachment force (FDF), which affects the berry stem's ability to support the weight of the berries and environmental stress. However, the majority of the FDF measurement methodologies used in earlier studies rely on destructive methods, which not only preclude future studies on the same samples but also substantially raise experiment repeatability error. In this study, a nondestructive method was developed to predict FDF based on grape visible features, allowing the change in FDF to be observed at any point during the postharvest preservation of grapes. First, physiological indexes related to FDF were screened and subsequently, 10 highly correlated indexes, such as berry color, berry weight, berry length, etc., were obtained. Thereafter, four machine learning models such as multiple linear regression (MLR), principal component regression (PCR), back propagation (BP) neural networks and genetic algorithm back propagation (GA-BP) neural networks were employed to predict FDF from relatively highly correlated physiological indexes. The results suggested that GA-BP model had the highest prediction efficiency with the correlation coefficient (R2), root mean square error (RMSE) and mean absolute percentage error (MAPE) of R2 = 0.833, RMSE = 0.426, MAPE = 0.163, respectively. Finally, the nondestructive FDF prediction model by the GA-BP model was developed using nondestructive apparent characteristics extracted using machine vision technology. This model achieved a good fitting effect, with R2 = 0.812, RMSE= 0.426, and MAPE= 0.334, respectively. In order to monitor the FDF change during grape postharvest storage and predict grape abscission, an effective and nondestructive FDF prediction method has been successfully developed. This encourages the studies on the physiological and molecular mechanism of abscission, and the use of precise fresh-keeping techniques for postharvest grape in the future.