<![CDATA[Tomato ripeness detection is an essential component in the development of automated agricultural systems, enabling improvements in harvesting accuracy, sorting consistency, and supply chain standardization. Conventional grading processes rely heavily on manual observation, which is subjective, labor-intensive, and unsuitable for large-scale operations. Recent advancements in deep learning enable automated recognition of visual maturity indicators through object detection frameworks, offering a more reliable and scalable solution. This study examines the implementation of two modern detection models, YOLO and DETR, for multi-level tomato ripeness classification involving four distinct maturity stages. The research workflow includes dataset collection, annotation, preprocessing, model training, threshold calibration, and systematic evaluation to assess detection stability and classification behavior under real-world variability.Analysis of model outputs demonstrates that both architectures are capable of identifying multiple ripeness stages with useful levels of consistency, although each model exhibits strengths under different operational conditions. YOLO provides advantages in scenarios requiring real-time responsiveness and deployment on resource-limited hardware, making it suitable for mobile automation and field-based harvesting systems. DETR shows improved interpretive behavior in visually complex environments, particularly when samples exhibit subtle maturity differences or appear in overlapping cluster formations. The findings indicate that no single model is universally optimal and that deployment choice should be based on application requirements, environmental constraints, and operational objectives. This research contributes practical insight to the integration of artificial intelligence in agriculture and provides a foundation for future work exploring model fusion, advanced feature learning, or multispectral input integration to further enhance maturity classification performance.]]>
