A Low-Cost Embedded Machine Learning-Based Acoustic Sensing Device for On-Tree Jackfruit Maturity Classification

This study presents a fully integrated machine learning–based acoustic sensing device for real-time, on-tree maturity classification of jackfruit, in which low-cost hardware design, embedded software, and an efficient machine learning model are integrated within a unified platform. On-site tapping acoustic signals were collected directly from fruits attached to the tree and transformed into frequency-domain features, including peak frequency, spectral centroid, and Mel-frequency cepstral coefficients. A total of 1,080 samples from three maturity stages (immature, mature, and ripened) were used to develop and evaluate multiple machine learning models. The results reveal distinct maturity-dependent frequency features, while the influence of tapping position was limited, demonstrating robustness under practical field conditions. Among the evaluated models, the Support Vector Machine achieved the best overall performance, with a testing accuracy of 92.3% and stable precision across maturity classes. When deployed on the proposed device, all models exhibited processing times of less than 1 second, confirming their suitability for real-time operation. With a compact handheld form factor and a total hardware cost of approximately USD 35, the proposed device provides a low-cost, practical, and scalable solution for on-site fruit maturity assessment, with strong potential for extension to other fruits and acoustic-based agricultural sensing applications.