A non‐destructive approach: Estimation of melon Fruit quality attributes and nutrients using hyperspectral imaging coupled with machine learning

Rapid and accurate biomass, nutrients, and sugar estimation facilitates efficient plant phenotyping and site-specific crop management. The hyperspectral technique enabled rapid and non-destructive determination of Nitrogen, Potassium and sucrose concentration in melon ( C.melo ) leaves and fruit using a spectral reflectance. The best modal for Nitrogen and Potassium quantitative prediction was precisely assessed using the MobileNet-v3-l model that exhibited the highest accuracy, with R² being 0.958, MSE 12.188 and MAE 1.519. Compared with the highest accuracy model of RGB R², MSE decreased by 21 %, and MAE decreased by 16 %. Meanwhile, the ResNet18 model has the highest accuracy, R² is 0.921, MSE is 16.246, and MAE is 1.851. Compared with the highest accuracy model of RGB R², MSE is increased by 85 %, and MAE is increased by 8 % in the Potassium model. In the sucrose model, the RegNet-y-8gf model had the highest accuracy, with R² of 0.958, MSE of 8.776 and MAE of 1.707. Furthermore, in the reductive sugar model, the accuracy of using RGB hyperspectral imaging ResNet18 model is the highest, R² is 0.936, MSE is 0.517, and MAE is 0.471. The present study shows the potential of the use of HSI technology directly in the field by proximal measurements under natural light conditions for the prediction of the harvest time of the melon.