Sensors for fruit firmness assessment: Comparision and fusion

Abstract

Non-destructive measurement of fruit firmness is a difficult problem and many different sensors have been developed in order to achieve this task. Three different European laboratories were associated in collaborative experiments on peaches, to compare three different sensing techniques, namely, sound, impact and micro-deformation. A Bayesian classifier is associated with each individual sensor and provides a classification into three categories, namely “soft”, “half firm” and “firm”. The fusion of the different sensors is performed by using Bayesian classifiers associated with heuristic methods for identity fusion. The result of the identity fusion is compared with the classification provided by an unsupervised algorithm based on destructive measurements. The fusion process provides some improvement in the classification results. For the individual sensors, the error rate of the classification varied from 19 to 28%, but the fusion process reduced this to 14%. Moreover, all measures of agreement between sensors lead to the conclusion that fusing sensors is better than using individual sensors