Rapid and Non‐Destructive Detection of Multiple Quality Attributes in Loquats During Storage Using Fiber Optic Spectroscopy and Machine Learning

Loquats are a valuable fruit, and assessing their quality is important for quality control and sorting. In this study, we developed a rapid, non‐destructive, novel method using fiber optic spectroscopy (200–1000 nm) and machine learning to determine multiple loquat quality attributes, including weight loss rate, pH, color index, carotenoids, firmness, and total flavonoid content. Competitive adaptive reweighted sampling (CARS), random frog, and successive projection algorithm were employed to extract feature variables and construct partial least squares regression (PLSR), multiple linear regression, and back propagation neural network models. To enhance model performance, variables from 1024 full spectra were further optimized via PLSR, yielding 17–30 optimal latent variables for six loquat quality attributes. The CARS‐PLSR model based on latent variables provided the most optimal results, with residual prediction deviation (RPD) of 4.398, 2.490, 3.500, 2.540, 2.172, and 2.287 for the six quality attributes. Therefore, this study offers a valuable reference for the rapid and non‐destructive detection of multiple quality attributes in fruits.