Non-Destructive Ultrasound Disorder and Disease Diagnostics for Harumanis Mango using Machine Learning and Deep Learning

Abstract

This study presents an innovative approach in post-harvest internal qualities diagnostics for Harumanis mango using ultrasound transducer combined with machine learning and deep learning techniques. Harumanis mango is highly susceptible to disorder and diseases. Conventional methods such as visual inspection, can be time-consuming and prone to inaccuracies. To address these challenges, this study investigates the use of ultrasound waves to capture internal structural variations in Harumanis mango. The acquired ultrasound data are subsequently processed using algorithms to classify the internal qualities of the fruit. One machine learning model and one deep learning model were trained and evaluated under a three-tier framework comprising held-out test accuracy, 5-fold cross-validation and prediction performance on 10 completely unseen mango samples. The Random Forest classifier achieved a test accuracy of 84%, a cross-validation accuracy of 82.4%. The One-Dimensional Convolutional Neural Network, trained achieved a test accuracy of 87% and a diseased-class recall of 92%. This research provides valuable insights into the advancement of precision agriculture, particularly in post-harvest quality assessment and internal qualities diagnostics. Future work may focus on refining model accuracy and adapting the system for field use under various environmental conditions.