Non-destructive identification of moldy walnut based on NIR

Walnuts are prone to developing mold during long-term freezing storage. Sampling shell breaking is usually used to detect the internal mold. In this paper, a prediction model of walnut mold was established by using near-infrared spectroscopy to realize the non-destructive detection of walnut mold. A combination of smooth, multiplicative scattering correction (MSC), and detrending was used as preprocessing method for the spectra. Competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), and principal component analysis (PCA) were used to extract the features from the 900–1700 nm band of the preprocessed spectra. Support vector machine (SVM) and extreme learning machine (ELM) models were developed, and both achieved 100% accuracy in identifying moldy walnuts. In addition, online identification system for moldy walnut was built to realize automatic spectra collection, moldy identification, and sort. The results show that the model developed can accurately predict moldy walnut. The online system established provides a viable solution for the online detection of moldy walnuts. The non-destructive nature of the method means that it can be applied without damaging the walnuts, which is an advantage over traditional shell-breaking methods.