Front-Face Fluorescence Spectroscopy and Feature Selection for Fruit Classification Based on N-CovSel Method

Internal disorder is a major problem in fruit production and is responsible for considerable economical losses. Symptoms are not externally visible, making it difficult to assess the problem. In recent years, 3D fluorescence spectroscopy has been used to reveal features of interest in agronomical field, such as plant stress and plant infection. Such technique could provide useful information regarding changes that occur at the tissue level, in order to distinguish spectral differences between healthy and disordered fruits. This paper introduces the use of the new three-way feature extraction N-CovSel method, compared to the commonly used N-PLS-DA method. These approaches were used upon front-face fluorescence spectra of 27 fruit pulp and skin samples, by analysing excitation wavelengths ranging from 250 to 650 nm, and emission wavelengths varying from 290 to 800 nm. N-CovSel method was applied to identify the most relevant features on: 1) excitation-emission wavelength couples, 2) excitation wavelengths whatever the emission wavelengths and 3) emission wavelengths whatever the excitation wavelengths. Discriminant analysis of the selected features were performed across classes. The constructed models provided key features to differentiate healthy fruits from disordered ones. These results highlighted the capability of N-CovSel method to extract the most fitted features for enhanced fruit classification using front-face fluorescence spectroscopy. They revealed characteristic fluorophores involved in the structural modifications generated by the physiological disorder studied. This paper provides preliminary results concerning the suitability of N-CovSel method for the desired application. Further investigations could be performed on intact fresh fruits in a non-destructive way, allowing an earlier and faster detection of the internal disorder for in-field or industrial applications.