Hyperspectral Imaging for Identification of Early Freeze Damage in Oranges

ABSTRACT The orange affected by freeze damage will experience the loss of partial nutrients, decline in taste quality, and acceleration of deterioration. According to the early signs of freeze damage, it becomes challenging for the naked eye to differentiate between oranges with freeze damage and those that are intact. This study aimed to investigate the feasibility of using hyperspectral imaging for identification of oranges with different degrees of early freeze damage. Firstly, the fundamental mechanism of hyperspectral imaging for identifying early freeze damage of orange was investigated by weight change before and after freeze damage, soluble solids content, and fluorescence images of the oranges with different degrees of early freeze damage. Subsequently, a variance inflation factor modified competitive adaptive reweighted sampling (VIF‐CARS) algorithm and a reptile search algorithm optimized support vector machine (RSA‐SVM) were proposed to extract effective wavelengths associated with freeze damage information of oranges and identify oranges with different degrees of early freeze damage. The results demonstrated that VIF‐CARS exhibited promising outcomes in dimensionality reduction, effective information extraction, and wavelength interpretability compared with traditional methods. Moreover, when combined with the RSA‐SVM model, this combination showcased reliable discrimination capabilities for identifying oranges with different degrees of early freeze damage, achieving the optimal classification accuracy of 94.44% and the F1‐score of 0.9484. Therefore, hyperspectral imaging offers a feasible approach for identification of oranges with different degrees of early freeze damage, which is a crucial step toward improving the detection system for the fruit supply chains.