Low-Cost, Multisensor Nondestructive Banana Ripeness Estimation Using Machine Learning

Ripeness is a key metric in the growth, distribution, and sale of fruits. Present methods to determine a given fruit’s ripeness rely on slow, labor intensive, and destructive methods that only provide an indication of crop-wide ripeness. A digital, low-cost, and nondestructive method would provide instant access to accurate ripeness data whenever required. In this article, a methodology is proposed for estimation of banana ripeness using an array of low-cost sensing modalities including cameras, spectrometers, volatile organic compound (VOC), and environmental sensors. Data were collected over five periods, each 35 days long (on average), using a setup designed for this systematic data collection. The banana’s ripeness is classified into one of five stages using a range of machine-learning (ML) algorithms such as support vector machines (SVMs), random forest (RF), K-nearest neighbors (KNN), gradient boosting classifiers (GBCs), and artificial neural networks (ANNs), which were trained and tested against datasets constructed from different combinations of sensor data. RF and GBCs were found to be the most accurate at 99.95% each when using data from all available sensors arrays.