MANGGA project targets to automate mango grading in Cebu

The manual classification of mangoes has long been a bottleneck in the mango supply chain, characterized by time-consuming efforts and subjective judgment. In response to this challenge, the “Mango Automated Neural Net Generic Grade Assignor (MANGGA)” project of the University of the Philippines Cebu (UP Cebu) harnesses the power of artificial intelligence (AI) and brings automation to the labor-intensive task of sorting Carabao mangoes for the fresh export market.

UP Cebu Professor Jonnifer Sinogaya leads this two-year project funded by the Philippine Council for Agriculture, Aquatic and Natural Resources Research and Development of the Department of Science and Technology (DOST-PCAARRD). The team has been working on the development of data acquisition system technologies for mangoes in cooperation with the Department of Agriculture-Region VII (DA-VII), University of the Philippines Los Baños-Postharvest Horticulture Training and Research Center (UPLB-PHTRC), and the Technological Institute of the Philippines (TIP).

The team’s systematic approach to data acquisition has led to an extensive data set of 10,440 images captured from various angles and orientations and corresponding ethylene concentrations collected from 870 individual mangoes, which served as the cornerstone for training a cutting-edge AI model for sorting Carabao mangoes.

The MANGGA project team has coded the Convolutional Neural Network (CNN) from scratch and also created an image data acquisition system. Their preliminary training of a single-input CNN model exhibited an impressive 94% accuracy in determining whether mangoes are suitable for export based on their overall visual characteristics.

Using the Philippine National Standard for quality metrics, the refinement of the CNN and Computer Vision System (CVS) promises a more efficient way to grade export-quality Carabao mangoes.

Midway through its second year, the project team refines its approach and explores innovative preprocessing techniques. The project also aims to continually assess multi-input CNN models and image data acquisition systems to achieve a higher level of precision.

The MANGGA project encourages the adoption of smart postharvest system within the local mango industry. With the premise of creating a conveyor system designed to sort mangoes based on their marketability, this initiative stands poised to revolutionize mango grading, offering efficiency and safety to the fresh export market. (Thea Mariel Valdeavilla, DOST-PCAARRD S&T Media Services)