Namdokmai Mango Sweetness Classification Using YOLO with Quality Focal Loss

The rising demand for fresh, high-quality fruits drives collaboration between agriculture and technology sectors. In Thailand, this demand extends to the popular Namdokmai mango, prized for its unique sweet-tart flavor. Accurately assessing Namdokmai mango sweetness is crucial for efficient fruit production and optimal consumer satisfaction. However, traditional methods lack accuracy, leaving consumers to gauge sweetness only after consumption. This study addresses this challenge by developing the You Only Look Once, or YOLO, a popular object detection and image segmentation model for its high speed and accuracy with the Quality Focal Loss (QFL), a variant of focal loss designed to balance classification and localization accuracy by assigning higher weights to high-quality predictions, enhancing the precision of object detectors. The model is trained on images of Namdokmai mangoes to predict their sweetness levels. By harnessing the ability of the YOLO to recognize complex patterns in images, this study aims to significantly improve the accuracy and efficiency of Namdokmai mango sweetness identification. As a result, YOLOv10 medium size (YOLOv10m) with QFL archives mean Average Precision (mAP) of 97% and macro F1-score of 63%. This not only benefits producers by optimizing the sorting and grading process but also empowers consumers to make informed choices based on their preferred sweetness level. The successful implementation of model can revolutionize the Namdokmai mango industry and paving the way for more profitable.