Banana Ripeness Estimation Using a Non-Destructive Approach Composed of an Array of Multimodal Sensors and Machine Learning

Ripeness is a crucial metric in the cultivation and sale of fruits. Current techniques for determining a fruit's ripeness rely on damaging, labour-intensive methods, providing only an approximation of crop-wide ripeness. A cost-effective, non-destructive ripeness detection method could provide instantaneous and accurate ripeness data whenever required, without generating wastage. This paper proposes a methodology for the measurement and estimation of banana ripeness using an array of low-cost sensing modalities. A 5-stage classification framework is proposed to classify bananas using a number of Machine Learning algorithms such as Support Vector Machines, Random Forest, K-Nearest Neighbours, Gradient Boosting Classifiers & Artificial Neural Networks. Datasets constructed from different combinations of sensor data were used for training and testing processes. The most accurate methods were Random Forest and Gradient Boosting Classifiers at 99.95 % each when using data from all available sensors.