YOLO-Lychee: A YOLOv8s-Based Detector for Lychee Growth-Stage Detection under Natural Orchard Conditions
DOI:
https://doi.org/10.4108/airo.12140Keywords:
Precision agriculture, Deep learning, Object detection, Lychee growth-stage detection, YOLOv8Abstract
Accurate lychee growth-stage detection in natural orchards is challenging because blossoms and fruits are often small, densely clustered, partially occluded, and visually similar to surrounding foliage. This study proposes YOLO-Lychee, a YOLOv8s-based detector for four developmental stages: blossom, young fruit, green fruit, and ripe fruit. The dataset contains 1,145 original orchard images and 19,422 annotated instances collected in Hai Duong and Bac Giang provinces, Vietnam. To reduce blossom-class imbalance, exposure-based augmentation was applied only to blossom samples in the training subset, resulting in 1,324 images and 20,261 annotations, while the validation and test sets remained unchanged. YOLO-Lychee replaces the original SPPF module with a Spatial and Channel Cross-Transformer module, incorporates a Context Augmentation Module in the neck, and uses EIoU instead of the default CIoU loss for bounding-box regression. Across six random seeds, YOLO-Lychee achieved a Precision of 83.02 ± 2.38%, Recall of 82.19 ± 1.87%, mAP50 of 88.42 ± 0.38%, and mAP50:95 of 72.76 ± 0.45%. Compared with recent YOLO-family detectors under the same protocol, the proposed model obtained the highest mAP50 and competitive mAP50:95 while maintaining real-time inference. Ablation results confirm the complementary contributions of SC3T, CAM, and EIoU, whereas qualitative analysis shows that blossom detection remains the most challenging case. These results demonstrate that YOLO-Lychee is a practical empirical baseline for lychee growth-stage detection and vision-assisted orchard monitoring.
Downloads
References
[1] NV Dung, DQ Nghi, NQ Hung, and NQ Huy. Currentproduction and development trends of lychee (litchichinensis sonn.) in vietnam in the coming future. InVI International Symposium on Lychee, Longan and OtherSapindaceae Fruits, pages 7–14, 2019.
[2] Naoshi Kondo, Mitsuji Monta, and Noboru Noguchi.Agricultural robots: mechanisms and practice. ApolloBooks, 2011.
[3] Inkyu Sa, Zongyuan Ge, Feras Dayoub, Ben Upcroft,Tristan Perez, and Chris McCool. Deepfruits: A fruitdetection system using deep neural networks. sensors,16(8):1222, 2016.
[4] Aleana Gongal, Suraj Amatya, Manoj Karkee, QinZhang, and Karen Lewis. Sensors and systems for fruitdetection and localization: A review. Computers andElectronics in Agriculture, 116:8–19, 2015.
[5] Ata Jahangir Moshayedi, Amir Sohail Khan, YiguoYang, Jiandong Hu, and Amin Kolahdooz. Robots inagriculture: Revolutionizing farming practices. EAIEndorsed Transactions on AI and Robotics, 3, 2024.
[6] Ameer Tamoor Khan, Sign Marie Jensen, and NomanKhan. Advancing food security through precisionagriculture: Yolov8’s role in efficient pest detection andmanagement. EAI Endorsed Transactions on AI andRobotics, 4, 2025.
[7] Ata Jahangir Moshayedi, Amir Sohail Khan, AminKolahdooz, Aiman Elragig, Zeashan Khan, and DavidBassir. Smart agro-ecosystem: A review of llm-basedrobotic systems for sensing and decision support inprecision agriculture. EAI Endorsed Transactions on AIand Robotics, 5, 2026.
[8] Ross Girshick. Fast r-cnn. In Proceedings of the IEEEinternational conference on computer vision, pages 1440–1448, 2015.
[9] Subramanian Parvathi and Sankar Tamil Selvi. Detectionof maturity stages of coconuts in complex backgroundusing faster r-cnn model. biosystems engineering,202:119–132, 2021.
[10] Suchet Bargoti and James Underwood. Deep fruitdetection in orchards. In 2017 IEEE internationalconference on robotics and automation (ICRA), pages3626–3633. IEEE, 2017.
[11] Juan Pablo Vasconez, Jose Delpiano, Stavros Vougioukas,and F Auat Cheein. Comparison of convolutionalneural networks in fruit detection and counting: Acomprehensive evaluation. Computers and Electronics inAgriculture, 173:105348, 2020.
[12] Hamzeh Mirhaji, Mohsen Soleymani, Abbas Asakereh,and Saman Abdanan Mehdizadeh. Fruit detection andload estimation of an orange orchard using the yolomodels through simple approaches in different imagingand illumination conditions. Computers and Electronicsin Agriculture, 191:106533, 2021.
[13] Craig B MacEachern, Travis J Esau, ArnoldWSchumann,Patrick J Hennessy, and Qamar U Zaman. Detectionof fruit maturity stage and yield estimation in wildblueberry using deep learning convolutional neuralnetworks. Smart Agricultural Technology, 3:100099, 2023.
[14] Lei Shen, Jinya Su, Runtian He, Lijie Song, Rong Huang,Yulin Fang, Yuyang Song, and Baofeng Su. Real-timetracking and counting of grape clusters in the fieldbased on channel pruning with yolov5s. Computers andElectronics in Agriculture, 206:107662, 2023.
[15] Dandan Wang and Dongjian He. Channel pruned yolov5s-based deep learning approach for rapid and accurateapple fruitlet detection before fruit thinning. BiosystemsEngineering, 210:271–281, 2021.
[16] G. Jocher, A. Chaurasia, and J. Qiu. Yolo by ultralytics. https://github.com/ultralytics/ultralytics,2023.
[17] Chien-Yao Wang, I-Hau Yeh, and Hong-Yuan MarkLiao. Yolov9: Learning what you want to learn usingprogrammable gradient information. arXiv preprintarXiv:2402.13616, 2024.
[18] Ao Wang, Hui Chen, Lihao Liu, Kai Chen, ZijiaLin, Jungong Han, and Guiguang Ding. Yolov10:Real-time end-to-end object detection. arXiv preprintarXiv:2405.14458, 2024.
[19] Nidhal Jegham, Chan Young Koh, Marwan Abdelatti,and Abdeltawab Hendawi. Evaluating the evolutionof yolo (you only look once) models: A comprehensivebenchmark study of yolo11 and its predecessors. arXivpreprint arXiv:2411.00201, 2024.
[20] Yunjie Tian, Qixiang Ye, and David Doermann. Yolov12:Attention-centric real-time object detectors. arXivpreprint arXiv:2502.12524, 2025.
[21] Glenn Jocher, Jing Qiu, Mengyu Liu, Shuai Lyu,Fatih Cagatay Akyon, and Muhammet Esat Kalfaoglu.Ultralytics yolo26: Unified real-time end-to-end visionmodels. arXiv preprint arXiv:2606.03748, 2026.
[22] Baoling Ma, Zhixin Hua, Yuchen Wen, Hongxing Deng,Yongjie Zhao, Liuru Pu, and Huaibo Song. Usingan improved lightweight yolov8 model for real-timedetection of multi-stage apple fruit in complex orchardenvironments. Artificial Intelligence in Agriculture,11:70–82, 2024.
[23] Aobin Zhu, Ruirui Zhang, Linhuan Zhang, TongchuanYi, Liwan Wang, Danzhu Zhang, and Liping Chen.Yolov5s-cedb: A robust and efficiency camellia oleiferafruit detection algorithm in complex natural scenes.Computers and Electronics in Agriculture, 221:108984,2024.
[24] Wenbai Chen, Mengchen Liu, ChunJiang Zhao, XingxuLi, and Yiqun Wang. Mtd-yolo: Multi-task deepconvolutional neural network for cherry tomato fruitbunch maturity detection. Computers and Electronics inAgriculture, 216:108533, 2024.
[25] Jiuxin Wang, Man Liu, Yurong Du, Minghu Zhao,Hanlang Jia, Zhou Guo, Yaoheng Su, Dingze Lu, andYucheng Liu. Pg-yolo: An efficient detection algorithmfor pomegranate before fruit thinning. EngineeringApplications of Artificial Intelligence, 134:108700, 2024.
[26] Fan Meng, Jinhui Li, Yunqi Zhang, Shaojun Qi, andYunchao Tang. Transforming unmanned pineapplepicking with spatio-temporal convolutional neuralnetworks. Computers and Electronics in Agriculture,214:108298, 2023.
[27] Xueyan Zhu, Fengjun Chen, Xinwei Zhang, Yili Zheng,Xiaodan Peng, and Chuang Chen. Detection the maturity of multi-cultivar olive fruit in orchard environmentsbased on olive-efficientdet. Scientia Horticulturae,324:112607, 2024.
[28] Tian Zhang, Dongfang Zhao, Yesheng Chen, HongliZhang, and Shulin Liu. Deepsort with siameseconvolution autoencoder embedded for honey peachyoung fruit multiple object tracking. Computers andElectronics in Agriculture, 217:108583, 2024.
[29] Thiago T Santos, Kleber XS de Souza, Joao CamargoNeto, Luciano V Koenigkan, Alecio S Moreira, and SoniaTernes. Multiple orange detection and tracking with 3-dfruit relocalization and neural-net based yield regressionin commercial sweet orange orchards. Computers andElectronics in Agriculture, 224:109199, 2024.
[30] Arturo Aquino, Juan Manuel Ponce, and Jose ManuelAndujar. Identification of olive fruit, in intensiveolive orchards, by means of its morphological structureusing convolutional neural networks. Computers andElectronics in Agriculture, 176:105616, 2020.
[31] Hatice Catal Reis and Veysel Turk. Potato leaf diseasedetection with a novel deep learning model based ondepthwise separable convolution and transformer networks.Engineering Applications of Artificial Intelligence,133:108307, 2024.
[32] Joao Mendes, Jose Lima, Lino Costa, Nuno Rodrigues,and Ana I Pereira. Deep learning networks forolive cultivar identification: A comprehensive analysisof convolutional neural networks. Smart AgriculturalTechnology, 8:100470, 2024.
[33] Hai-Binh Le, Thai Dinh Kim, Manh-Hung Ha, AnhLong Quang Tran, Duy-Thuc Nguyen, and Xuan-MinhDinh. Robust surgical tool detection in laparoscopicsurgery using yolov8 model. In 2023 InternationalConference on System Science and Engineering (ICSSE),pages 537–542. IEEE, 2023.
[34] Zhaohui Zheng, Ping Wang, Wei Liu, Jinze Li, RongguangYe, and Dongwei Ren. Distance-iou loss: Fasterand better learning for bounding box regression. InProceedings of the AAAI conference on artificial intelligence,volume 34, pages 12993–13000, 2020.
[35] Manh-Tuan Do, Manh-Hung Ha, Duc-Chinh Nguyen,Kim Thai, and Quang-Huy Do Ba. Human detectionbased yolo backbones-transformer in uavs. In2023 International Conference on System Science andEngineering (ICSSE), pages 576–580. IEEE, 2023.
[36] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and JianSun. Spatial pyramid pooling in deep convolutionalnetworks for visual recognition. IEEE transactions onpattern analysis and machine intelligence, 37(9):1904–1916, 2015.
[37] Jinsheng Xiao, Tao Zhao, Yuntao Yao, Qiuze Yu, and YunhuaChen. Context augmentation and feature refinementnetwork for tiny object detection, 2022. Retrieved from https://openreview.net/forum?id=q2ZaVU6bEsT.
[38] Yi-Fan Zhang, Weiqiang Ren, Zhang Zhang, Zhen Jia,Liang Wang, and Tieniu Tan. Focal and efficient iou lossfor accurate bounding box regression. Neurocomputing,506:146–157, 2022.
[39] Hamid Rezatofighi, Nathan Tsoi, JunYoung Gwak, AmirSadeghian, Ian Reid, and Silvio Savarese. Generalizedintersection over union: A metric and a loss for bounding box regression. In Proceedings of the IEEE/CVF conferenceon computer vision and pattern recognition, pages 658–666, 2019.
[40] Zanjia Tong, Yuhang Chen, Zewei Xu, and Rong Yu.Wise-iou: Bounding box regression loss with dynamicfocusing mechanism. arXiv preprint arXiv:2301.10051,2023.
[41] Zhora Gevorgyan. Siou loss: More powerful learning forbounding box regression. arXiv preprint arXiv: 2205.12740, 2022.
Downloads
Published
Issue
Section
License
Copyright (c) 2025 Thi-Nguyen Nguyen, Manh-Tuan Do, Dinh-Thai Kim, Tuan-Minh Nguyen, Huyen-Trang Nguyen
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
This is an open access article distributed under the terms of the CC BY-NC-SA 4.0, which permits copying, redistributing, remixing, transformation, and building upon the material in any medium so long as the original work is properly cited.
