NLA: Natural Light Augmentation Method for Crop Image Datasets

Sangho Lee; Seong-Woo Jeon; Jayeong Paek; Manjung Kim; Sang-Hyeon Kang; Kyoung-Chul Kim; YoungKi Hong; Ki-Beom Lee

doi:10.12791/KSBEC.2025.34.4.588

All Issue

2025 Vol.34, Issue 4 Preview Page Next Page

Original Articles

NLA: Natural Light Augmentation Method for Crop Image Datasets NLA: 작물 이미지 데이터셋을 위한 자연광 기반 데이터 증강 기법

31 October 2025. pp. 588-595

PDF XML

Abstract

Data augmentation is a key factor for improving the performance of object detection models in computer vision. Agricultural imagery is captured in open fields and greenhouses, and lighting varies widely with shooting time, weather conditions, and the influence of structures. These characteristics destabilize the distribution of training data, and when conventional augmentations generate data that differ from real environments, they can degrade a model’s generalization. To address this problem, we propose Natural Light Augmentation (NLA), a lighting-aware augmentation that reflects illumination changes in agricultural images. NLA simulates variations in brightness, color temperature, and contrast so that conditions likely to occur in practice are applied to the training data. We validated the effectiveness of the proposed method on Oriental melon and strawberry datasets. On the Oriental melon dataset, NLA improved overall performance compared with existing augmentations, with the largest gain observed on mAP50-95 (4－6%). On the strawberry dataset, the Baseline model showed a decline in validation performance in the later training phase-suggesting overfitting-whereas NLA maintained stable performance as epochs progressed. We additionally conducted experiments with PatchSwap, which exchanges patches within the same image, but the performance improvement was limited. These results demonstrate that augmentation reflecting real lighting conditions in agricultural datasets effectively improves model performance and increases training stability. As an augmentation tailored to agricultural imagery, NLA should be extended to a wider range of crops and environmental conditions in future work.

Keywords

data augmentation

deep learning

object detection

oriental melon

strawberry

본 연구에서는 작물 이미지 데이터셋에 적합한 데이터 증강 기법으로 자연광 기반 증강을 제안하였다. NLA는 밝기, 색온도, 대비 변화를 반영하여 실제 환경에서 발생 가능한 광조건을 학습 데이터에 적용함으로써, 기존 YOLO 기본 증강 대비 더욱 현실적인 학습 데이터를 생성할 수 있도록 설계되었다. 참외와 딸기 데이터셋을 대상으로 한 실험 결과, NLA는 Baseline 대비 전반적인 성능 향상을 보였다. 또한 Baseline에서 학습 후반부에 나타난 성능 저하나 불안정성이 NLA에서는 완화되어, 제안 기법이 과적합 방지와 학습 안정성 유지에도 효과적임을 입증하였다. PatchSwap은 Mosaic의 한계를 보완하기 위한 보조적 시도로 실험에 포함되었으나, 종합적인 성능 향상에는 뚜렷한 기여를 하지 못하였다. 본 연구 결과는 작물 이미지에서 실제 광조건을 반영한 데이터 증강의 필요성과 유효성을 보여주었으며, 이는 향후 다양한 작물과 촬영 환경으로 확장 가능한 가능성을 제시한다. 앞으로는 NLA의 적용 범위를 확대하고, 다른 딥러닝 모델 및 추가적인 농업 데이터셋에 대한 성능 검증을 통해 본 기법의 범용성과 실효성을 더욱 강화할 필요가 있다.

키워드

객체 탐지

데이터 증강

딥러닝

딸기

참외

References

Bochkovskiy A., C.Y. Wang, and H.Y.M. Liao 2020, YOLOv4: Optimal speed and accuracy of object detection. arXiv:2004.10934. doi:10.48550/arXiv.2004.10934

10.48550/arXiv.2004.10934

Buslaev A., V. I. Iglovikov, E. Khvedchenya, A. Parinov, M. Druzhinin, and A. A. Kalinin 2020, Albumentations: Fast and flexible image augmentations. Information, 11:125. doi:10.3390/info11020125

10.3390/info11020125

Chen P., S. Liu, H. Zhao, X. Wang, and J. Jia 2020, GridMask data augmentation. arXiv:2001.04086. doi:10.48550/arXiv.2001.04086

10.48550/arXiv.2001.04086

Fan X., C. Ge, X. Yang, and W. Wang 2024, Cross-modal feature fusion for field weed mapping using RGB and near-infrared imagery. Agriculture 14:2331. doi:10.3390/agriculture14122331

10.3390/agriculture14122331

Ghiasi G., Y. Cui, A. Srinivas, R. Qian, T.Y. Lin, E. D. Cubuk, Q. V. Le, and B. Zoph 2021, Simple copy-paste is a strong data augmentation method for instance segmentation. CVPR. doi:10.1109/CVPR46437.2021.00294

10.1109/CVPR46437.2021.00294

Kaur R., U. Mittal, A. Wadhawan, A. Almogren, J. Singla, S. Bharany, S. Hussen, A. U. Rehman, and A.A. Al-Huqail 2025, YOLO-LeafNet: A robust deep learning framework for multispecies plant disease detection with data augmentation. Sci Rep 15:28513. doi:10.1038/s41598-025-14021-z

10.1038/s41598-025-14021-z40764650PMC12325615

Koirala A., Z. Wang, K.B. Walsh, and C. McCarthy 2019, Deep learning-Method overview and review of use for fruit detection and yield estimation. Comput Electron Agric 162:219-234. doi:10.1016/j.compag.2019.04.017

10.1016/j.compag.2019.04.017

Lee S., and S. Lee 2025, Efficient data augmentation methods for crop disease recognition in sustainable environmental systems. Big Data and Cognitive Computing 9:8. doi:10.3390/bdcc9010008

10.3390/bdcc9010008

Lee S.H., S.M. Woo, J.H. Choi, and J.O. Kim (2017, September). Two-step multi-illuminant color constancy for outdoor scenes. In 2017 IEEE International Conference on Image Processing (ICIP) (pp. 710-714). IEEE. doi:10.1109/ICIP.2017.8296373

10.1109/ICIP.2017.8296373

Lin, T.Y., M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollár, and C.L. Zitnick 2014, Microsoft COCO: Common objects in context. ECCV. doi:10.1007/978-3-319-10602-1_48

10.1007/978-3-319-10602-1_48

Liu, W., D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.Y. Fu, and A. C. Berg 2016, SSD: Single shot multibox detector. ECCV. doi:10.1007/978-3-319-46448-0_2

10.1007/978-3-319-46448-0_2

Redmon, J., and A. Farhadi 2018, YOLOv3: An incremental improvement. arXiv:1804.02767. doi:10.48550/arXiv.1804.02767

10.48550/arXiv.1804.02767

Reinhard, E., M. Stark, P. Shirley, and J. Ferwerda 2002, Photographic tone reproduction for digital images. SIGGRAPH. doi:10.1145/566570.566575

10.1145/566570.566575

Sa, I., Z. Ge, F. Dayoub, B. Upcroft, T. Perez, and C. McCool 2016, DeepFruits: A fruit detection system using deep neural networks. Sensors, 16:1222. doi:10.3390/s16081222

10.3390/s1608122227527168PMC5017387

Wang, C.Y., A. Bochkovskiy, and H.Y.M. Liao 2023, YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. CVPR. doi:10.1109/CVPR52729.2023.00721

10.1109/CVPR52729.2023.00721

Yun, S., D. Han, S.J. Oh, S. Chun, J. Choe, and Y. Yoo 2019, CutMix: Regularization strategy to train strong classifiers with localizable features. ICCV. doi:10.1109/ICCV.2019.00612

10.1109/ICCV.2019.00612

Information

Publisher :The Korean Society for Bio-Environment Control
Publisher(Ko) :(사)한국생물환경조절학회
Journal Title :Journal of Bio-Environment Control
Journal Title(Ko) :생물환경조절학회지
Volume : 34
No :4
Pages :588-595
Received Date : 2025-10-02
Revised Date : 2025-10-23
Accepted Date : 2025-10-30
DOI :https://doi.org/10.12791/KSBEC.2025.34.4.588

Journal of Bio-Environment Control ISSN:1229-4675(Print) 2765-3641(Online) 생물환경조절학회지

All Issue