November 06, 2024 | Agricultural and Biological Engineering |
Researchers from the Agricultural Engineering Research Institute and Khon Kaen University in Thailand conducted a study to address quality sorting challenges in mangosteen, a key economic crop. Mangosteen quality is often determined by skin color, which currently relies heavily on manual inspection by experts, particularly for fruits with damaged skin that are unsuitable for export. The study aimed to develop a more efficient and objective approach using image processing technology.
A total of 60 mangosteen fruits, representing six maturity levels, were imaged from four angles using RGB cameras, resulting in 480 images. These images were analyzed to distinguish between good and damaged skin using three machine learning algorithms: Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Decision Tree (Fine Tree). The algorithms showed comparable performance, achieving high average accuracy rates across maturity levels, ranging from 93.03% to 100%. However, the effectiveness was lower for levels 2 and 3 due to the similar appearance of good and damaged skin at these stages.
The findings demonstrate that image processing can reliably classify damaged-skinned mangosteen, providing a promising solution for automated quality sorting. This technology could improve efficiency and consistency in the mangosteen supply chain, benefiting producers and exporters.
