Deteksi kerusakan mekanis stroberi (Fragaria x ananassa) Menggunakan Citra RGB dan Fluoresen Berbasis Deep Learning

Abstract

Penelitian ini berhasil mengembangkan sistem deteksi kerusakan mekanis stroberi (memar dan sayat) berbasis deep learning menggunakan YOLOv11 (instance segmentation) pada dua model citra, RGB dan fluoresen. Evaluasi menunjukkan model fluoresen unggul pada seluruh metrik: akurasi 96,8%, recall 95,2%, precision 95,6%, dan F1-score rata-rata 93,0%; sedangkan model RGB mencapai akurasi 89,9%, recall 85,2%, precision 89,2%, dan F1-score rata-rata 82,8%. Model fluoresen unggul karena menghasilkan kontras visual yang lebih tajam. sehingga lebih efektif dalam membedakan area rusak (memar dan sayat) dari jaringan sehat, dengan potensi penerapan nondestruktif untuk inspeksi mutu. Meskipun demikian, sistem masih memiliki keterbatasan dalam mendeteksi kerusakan halus/berkontras rendah, yang tercermin dari false negative pada kelas memar dan sayat. Secara keseluruhan, pendekatan fluoresen YOLOv11 berpotensi diterapkan untuk inspeksi mutu pascapanen secara cepat dan nondestruktif.

This study successfully developed a deep learning-based strawberry mechanical damage detection system (bruises and cuts) using YOLOv11 (instance segmentation) on two image models, RGB and fluorescent. Evaluation showed that the fluorescent model excelled in all metrics: 96.8% accuracy, 95.2% recall, 95.6% precision, and 93.0% average F1-score; while the RGB model achieved 89.9% accuracy, 85.2% recall, 89.2% precision, and 82.8% average F1-score. The fluorescent model excelled due to its sharper visual contrast, making it more effective in distinguishing damaged areas (bruises and cuts) from healthy tissue, with potential as a nondestructive standard for quality inspection. However, the system still has limitations in detecting subtle/low-contrast damage, which is reflected in false negatives in the bruise and cut classes. Overall, the YOLOv11 fluorescent approach has potential for rapid and nondestructive postharvest quality inspection.