Development of an Optical Handheld Device and Decision Tree for Mango Ripeness Classification Based on Sweetness Quality

Abstract

Reducing postharvest losses and ensuring the quality of mango production can be achieved through accurate, rapid assessment of mango ripeness. Fruit incision and pulp analysis using chemical substances are among the traditional methods that are correlated to invasive, damaging, and impractical for widespread use. Non-destructive testing using non-invasive sensing mechanisms offers a feasible alternative. However, these existing technologies often require complex analysis and extensive calibration, thereby limiting portability and accessibility. This study presents the development of an economical handheld optical device that integrates optical sensors with a decision-tree machine learning algorithm to classify mango ripeness based on sweetness quality. The system employs a 470 nm LED in conjunction with a BPW34 photodiode to measure the light reflectance of the mango. Using MATLAB, the corresponding output, a voltage signal, and the ground-truth data (ripe and unripe status) from the calibration study are used as the input and output of the decision tree classifier, respectively.  The results yield an overall accuracy of 81.3%, whereas the benchmarking shows 100% correct ripeness status. The developed prototype provides a cost-effective, scalable, and user-friendly solution for precision agriculture, with recommendations for future enhancements in other internal quality aspects and large-scale benchmarking.