Banana spoilage benchmark determination method and early warning potential based on hyperspectral data during its storage

In order to realize the early-warning of banana spoilage during its storage using hyperspectral information, a method of banana spoilage benchmark is proposed, and on this basis the early-warning potential based on Mahalanobis distance (MD) is explored. Firstly, Wilks Λ statistic coupled with principal component analysis was employed to extract nine feature wavelengths. Secondly, the sixth storage day was preliminarily judged as the banana spoilage benchmark by the inflection point of the Voigt fitting curves of chromatism indexes. Meanwhile, the minimum values of all average reflection spectral curves at the 9 feature wavelengths were found to be the sixth storage day. And the sixth storage day was also found to be the cut-off point of these feature spectral grayscale intensities. Thus the spoilage benchmark was determined as the sixth storage day and represented by corresponding feature spectral information. Finally, the MD between the spoilage benchmark and the banana samples with different storage date was analyzed. The results showed that the MD was smaller and smaller with increasing storage date. This is consistent with the actual spoilage process of banana. Therefore, the results provide a feasible method for the early warning of banana spoilage.