ABSTRACT
ABSTRACT: A major issue in the post-harvest phase of the fruit production sector is the systematic determination of the maturity level of fruits, such as the ripeness of watermelons. Maturity assessment plays an important role while sorting in packing houses during the export. This paper proposes a support vector machine-based method for the automated non-destructive classification of watermelon ripeness by acoustic analysis. Acoustic samples are collected from ripe and unripe watermelons in a studio environment by thumping on the surface of watermelons. Sound samples are pre-processed to remove silence regions by fixing an energy threshold. Pre-processed sound signals are segmented into equal-length frames sized 200 ms, and Teager Energy Operator (TEO)—based features are extracted. The entire set of audio samples are divided into a training set with 60% of the total audio samples and the remaining 40% for testing. A support vector machine-based classifier is trained with features extracted from the training set. Twenty dimensional feature vectors are computed in the feature extraction phase and fed into the classification phase. The results show that the proposed TEO-based method was able to discriminate between ripe and unripe watermelons with overall accuracy of 83.35%.
