Earticle

초록

영어

In coffee cultivation, accurately classifying fruit ripeness to ensure harvest quality is an important task. In particular, traditional manual harvesting methods in smart farm environments are labor-intensive and prone to errors. In this study, we propose an improved YOLOv11-based system for automatically detecting coffee fruit ripeness in hydroponic environments. The proposed approach consists of two parts: class imbalance mitigation through Instance-Aware Repeat Factor Sampling (IRFS) and network architecture optimization. IRFS improves the sampling weight of rare classes by simultaneously considering class distributions at both the image level and instance level, overcoming the limitations of existing Repeat Factor Sampling, which only considers image-level frequencies. The network architecture uses EfficientNetV2 with MBConvolution blocks as the backbone for improved feature extraction. The neck combines the C2PSA block, which integrates the Cross Stage Partial with Partial Self-Attention mechanism, and the head combines the Dynamic Adaptation Head for adaptive detection. An RGB image dataset was collected from a smart farm in Siheung, Gyeonggi Province, South Korea, and consists of four maturity classes: unripe, semi_ripe, ripe, and overripe. The proposed system achieves 95% on the mAP@0.5 metric, demonstrating superior performance compared to existing YOLO models. Compared to existing models without IRFS, it shows a 5% improvement in performance compared to YOLOv11n. Therefore, consistent performance improvements are observed when applying IRFS to all YOLO models. These results indicate that the combination of class imbalance mitigation and an optimized network architecture is effective for accurate coffee berry maturity detection.

목차

Abstract
1. Introduction
2. Related Work
3. Proposed Method
3.1 Data Augmentation
3.2 Class Imbalance Mitigation Strategy
3.3 Performance Enhancement Network Design
3.4 Loss Function
4. Experimental Environment and Results
4.1 Experimental Environment
4.2 Experimental Results
5. Conclusion
Acknowledgement
References

키워드

저자

• Taewook Kim [ M.S, Department of Information Convergence System, Kwangwoon University, Seoul, South Korea ]
• Heejun Youn [ Doctoral Course, Department of Plasma Bio Display, Kwangwoon University, Seoul, South Korea ]
• Yuseong Lee [ M.S, Department of Information Convergence System, Kwangwoon University, Seoul, South Korea ]
• Yongcheon Cho [ Minong Bio Co., Seoul, South Korea ]
• Jin Sik Min [ Minong Bio Co., Seoul, South Korea ]
• Seunghyun Lee [ Professor, Ingenium College Liberal Arts, Kwangwoon University, Seoul, South Kore ]
• Soonchul Kwon [ Associate professor, Graduate School of Smart Convergence, Kwangwoon University, Seoul, South Korea ] Corresponding Author

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  International Journal of Internet, Broadcasting and Communication

• 간기

  계간

• pISSN

  2288-4920

• eISSN

  2288-4939

• 수록기간

  2009~2025

• 십진분류

  KDC 326 DDC 380