<![CDATA[Melon is a horticultural commodity with high economic value, and characteristics such as sweetness, aroma, texture, and phytonutrient content significantly influenced consumer preference. Conventional methods for determining melon ripeness were time-consuming, required considerable expertise, and were often prone to subjective errors, ultimately slowing the production and distribution process. This study aimed to detect the ripeness level of golden melon fruit non-destructively using the YOLOv11 algorithm, focusing on external physical characteristics as the basis for classification. The objectives included applying transfer learning to categorize golden melon into ripe and unripe classes and evaluating model performance using precision, recall, mAP50, mAP50-95, and F1-score. The research methodology consisted of a literature review, dataset collection from previous studies, system design, implementation, and performance testing. The dataset was divided into 70% training, 20% validation, and 10% testing data, and the Adam optimizer was used during the training phase. Based on four experimental scenarios, scenario 3 produced the best and most consistent results, achieving a precision of 90.58%, a recall of 90.79%, an mAP50 of 97.31%, an mAP50-95 of 88.84%, and an F1-score of 92.97%. These findings demonstrated that scenario 3 offered optimal performance for detecting golden melon ripeness. Thus, the model was highly reliable overall.]]>
Copyrights © 2025
