A LOW-COST ACOUSTIC SIGNAL PROCESSING APPROACH FOR DURIAN RIPENESS CLASSIFICATION USING WAVELET TRANSFORM AND FOURIER TRANSFORM

Abstract

Accurate determination of fruit ripeness, particularly for durian — a high-value fruit with a short post-harvest shelf life—is crucial in modern agricultural supply chains. This study proposes a method for classifying durian ripeness levels based on acoustic signals generated by tapping the fruit’s shell. The approach consists of three main stages: noise filtering and feature extraction using the Discrete Wavelet Transform (DWT), frequency-domain power spectrum analysis via the Fourier Transform, and computation of a spectral uncertainty index for classification. Experimental results on 75 samples from three ripeness groups (unripe, moderately ripe, overripe) show that “unripe” and “ripe” groups exhibit similar time-domain characteristics but are clearly distinguishable in terms of spectral uncertainty. With a classification threshold set at 33.35, the proposed method achieved 100% accuracy for the two primary classes. This approach does not require deep learning models, offers low implementation costs, and has strong potential for integration into intelligent agricultural sorting systems.