<![CDATA[Assessing sugarcane quality is crucial for ensuring both economic value and processing efficiency in sugar production. Conventional approaches, such as refractometer-based Brix measurements, are destructive, labor-intensive, and unsuitable for large-scale or rapid field evaluations. This highlights the need for non-destructive, automated solutions that can deliver accurate and scalable assessments. This study proposes a deep learning framework for classifying sugarcane internodes into two quality categories based on Brix values: unsuitable for milling (<16 °Brix) and suitable for milling (≥16 °Brix) using image-based analysis. The dataset consists of two configurations: Luar1 (single internode) and Luar2 (a split internode with two outer sides placed side by side), each photographed against white and black backgrounds. Preprocessing, data augmentation, and transfer learning were applied using VGG19 and ResNet50 under a two-phase strategy. Phase 1 involved freezing the backbone layers (50 epochs), and Phase 2 involved fine-tuning (100 epochs). The results demonstrate that fine-tuning significantly enhanced model performance. VGG19 achieved accuracies between 72.12% and 75.06%, while ResNet50 consistently outperformed it, reaching 78.85% with the Luar2_Putih dataset. Confusion matrix analysis further confirmed ResNet50’s superior ability to minimize misclassification, particularly for high-quality canes that are crucial for milling feasibility. These findings advance non-destructive quality assessment in sugarcane and support the United Nations Sustainable Development Goals (SDG 2, SDG 9, and SDG 12) by strengthening food security through improved crop utilization, fostering innovation in agricultural technologies, and promoting sustainable production practices in the sugar industry.]]>
