Evaluation of Ente plum maturity by infrared spectroscopy

For several years, B.I.P. team has conducted measurements on Ente plums in the aim of better characterizing the fruit maturity at harvest. The final objective is to be able to adapt the transformation process from the plum to the prune in optimizing the drying step.
The quality parameters identified are the sugar content and the acidity which are both linked to the maturity state of the plum. These parameters are currently quantified by laboratory methods that require time and the destruction of the sample. These constraints are not adapted to real field applications.
Thus, the project aims at using fast and nondestructive methods to predict the sugar content and the acidity of plums. Among them, several instruments delivering one single measurement have been tested (calculation of maturity indices, volumetric mass,…). However, due to the complexity of the biological samples, a solution based on visible/near infrared spectroscopy which provides a rich multivariate signal seemed to be more adapted.

The experiment has been conducted by the B.I.P. for 2 years in 2013 and 2014. 2500 fruits have been collected during the maturation phase, on thirteen different orchards to include the biological variability. Each fruit has been scanned on both faces by the Labspec4 from ASDi (ville, USA), in the spectral range 350nm - 2500nm.
After spectral measurement, each plum has been analyzed for sugar content and acidity by the standard laboratory methods.
Ondalys has then conducted the data analysis by means of chemometrics methods. Principal component analysis has been used to observe the influence factors such as orchard, location of the measurement and year. Then, Partial Least Squares Regression models have been implemented for the prediction of brix and acidity. The model robustness was assessed and optimized by applying different pre-processing techniques.

A first study has been conducted in 2013, assessing the ability of near infrared spectroscopy to predict plum quality. The models have been successfully developed with very good results for sugar and satisfactory ones for acidity. These models have then been tested on 2014 campaign in order to test the model robustness from one year to another one. Again, better results were obtained with sugar than with acidity, for which, a slope has been observed. Models have then been updated and re-optimized with both years.
Good performances have been obtained to evaluate the sugar content for each fruit. For acidity, the task was more complex. The model gives a satisfactory tendency on each fruit, and a very good estimation of the orchard averaged acidity at a given time.
The next step will be to finalize the study in validating models on a third campaign which will be conducted in November 2015.