基于 RSSRP_YOLOv8 的离体稻穗结实率检测方法

A RSSRP_YOLOv8-based method of detecting the rice seed-setting rate on detached rice panicle

摘要

[目的]针对水稻育种研究中离体稻穗样本结实率检测繁杂，脱粒检测易破坏稻穗结构等问题，本研究构建了一种基于改进YOLOv8s的稻穗结实率检测模型RSSRP_YOLOv8（rice seed-setting rate on panicle using YOLOv8）。[方法]借鉴轻量级跨尺度融合结构（cNN-based cross-scale feature fusion，ccff）重构原始模型颈部网络，减小模型体量及运算复杂性，提高检测效率；采用分组混洗卷积（grouped shuffle convolution，gSconv）取代模型骨干网络中的标准卷积和优化后PaN-fPN结构中的下采样标准卷积，促进稻穗籽粒特征信息提取与融合，提高识别准确率与效率；引入一种创新的动态上采样模块（dynamic sampling，dysample）取代传统采样模块，提高模型多尺度特征融合和处理能力，促进稻穗籽粒特征细节重建和复杂场景中的特征恢复。[结果]RSSRP_YOLOv8模型在测试集上的平均精度均值maP@0．5、精确度和召回率分别为98．2％、95．6％、94．2％，相比于原始模型分别提高了0．7％、0．6％、0．3％，模型浮点运算量和模型大小分别降低了26．2％和42．2％，模型检测帧率达到64．9f􀅰s^-1。在5个品种水稻共计1100张稻穗原始采集图上进行验证，模型对稻穗实粒、空秕粒和结实率的检测平均绝对误差（Mae）分别为2．09粒、1．77粒和0．92％，结实率平均检测精度达到99％以上；平均绝对百分比误差（MaPe）分别为2．27％、4．39％和1．42％。[结论]该方法可实现离体稻穗结实率无损检测，为水稻育种研究中离体稻穗样本结实率检测提供技术支撑。

Abstract

[Objectives] Detecting the seed-setting rate of detached rice panicles samples is a tedious laboratory work in rice breeding. Pre-treatment with threshing causes damage to rice panicle, thus challenging the reliability of the assessed results. Therefore, this study aimed to develop an improved YOLOv8s-based rice seed-setting rate detection model, i.e. rice seed-setting rate on panicle using YOLOv8 (RSSRP_YOLOv8). [Methods] The model was optimized by incorporating the concept of a lightweight CNN-based cross-scale feature fusion (CCFF) to restructure the original model's neck network, leading to reduced model size and computational complexity while improving detection efficiency. The grouped shuffle convolution (GSConv) was applied to replace standard convolutions in both the model's backbone and the optimized PAN-FPN structure's down-sampling layers, which enhanced the extraction and integration of rice grain features and improved recognition accuracy and efficiency. An innovative dynamic up-sampling module (Dysample) was introduced to replace traditional up-sampling methods, resulting in enhanced multi-scale feature fusion and processing capability and thereby improving the reconstruction of rice grain feature details and recovering features in complex scenes. [Results] The RSSRP_YOLOv8 model achieved a mean average precision (mAP@0.5), precision and recall of 98.2%, 95.6% and 94.2%, respectively, using the test set, and related improvements were 0.7%, 0.6%, and 0.3% as compared to the original model. The model's floating-point operations and size were reduced by 26.2% and 42.2%, respectively, as compared to the original model. The detection frame rate reached 64.9 f·s^-1. Validation on 1100 original rice panicle images from five rice varieties showed that the average absolute error (MAE) for detecting actual grains, empty grains, and seed setting rate were 2.09 grains, 1.77 grains, and 0.92%, respectively, with seed setting rate detection accuracy exceeding 99%. The mean absolute percentage error (MAPE) was 2.27%, 4.39% and 1.42%, respectively. [Conclusions] This method enables non-destructive detection of the seed-setting rate in detached rice panicles, providing technical support for seed-setting rate detection in detached rice panicles samples in rice breeding research.

关键词

水稻 / 结实率 / 图像识别 / YOLOv8 / 无损检测 / 表型

Key words

rice / seed-setting rate / image recognition / YOLOv8 / non-destructive detection / phenotype

参考文献

[1] 王勇, 陶兆胜, 石鑫宇, 等. 基于改进YOLOv5s的不同成熟度苹果目标检测方法[J]. 南京农业大学学报,2024,47(3): 602-611.

[2] Wang Y, Tao Z S, Shi X Y, et al. Apple target detection method with different ripeness based on improved YOLOv5s[J]. Journal of Nanjing Agricultural University, 2024, 47(3): 602-611 (in Chinese with English abstract).

[3] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// 2014 IEEE Conference on Computer Vision and Pattern Recognition. June 23-28, 2014, Columbus, OH, USA. IEEE, 2014: 580-587.

[4] Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). June 27-30, 2016, Las Vegas, NV, USA. IEEE, 2016: 779-788.

[5] 岳凯, 张鹏超, 王磊, 等. 基于改进YOLOv8n的复杂环境下柑橘识别[J]. 农业工程学报,2024,40(8): 152-158.

[6] Yue K, Zhang P C, Wang L, et al. Recognizing Citrus in complex environment using improved YOLOv8n[J]. Transactions of the Chinese Society of Agricultural Engineering, 2024, 40(8): 152-158 (in Chinese with English abstract).

[7] 曹丽英, 钟舸澳, 赵浩宇, 等. 一种基于无人机Rgb图像的玉米雄穗检测方法[J]. 沈阳农业大学学报,2025,56(6): 20-30.

[8] Cao L Y, Zhong G A, Zhao H Y, et al. A detecting method for maize tassels based on UAV RGB images[J]. Journal of Shenyang Agricultural University, 2025, 56(6): 20-30 (in Chinese with English abstract).

[9] 吴利刚, 陈乐, 刘泽鹏, 等. 基于YOLOv8-abW的黄花成熟度检测方法[J]. 农业工程学报,2024,40(13): 262-272.

[10] Wu L G, Chen L, Liu Z P, et al. YOLOv8-ABW based method for detecting Hemerocallis citrina Baroni maturity[J]. Transactions of the Chinese Society of Agricultural Engineering, 2024, 40(13): 262-272 (in Chinese with English abstract).

[11] 李瑞鑫, 曹悦, 司秀丽, 等. 基于改进YOLOv8n的樱桃成熟度检测研究[J]. 智能化农业装备学报(中英文),2025,6(4): 40-49.

[12] Li R X, Cao Y, Si X L, et al. Research on cherry ripeness detection based on improved YOLOv8n[J]. Journal of Intelligent Agricultural Mechanization, 2025, 6(4): 40-49 (in Chinese with English abstract).

[13] Lu Y, Yu J H, Zhu X F, et al. YOLOv8-Rice: a rice leaf disease detection model based on YOLOv8[J]. Paddy and Water Environment, 2024, 22(4): 695-710.

[14] 靳新宇, 于复兴, 索依娜, 等. 基于改进YOLOv8的水稻病害检测算法[J]. 江苏农业学报,2025,41(3):537-548.

[15] Jin X Y, Yu F X, Suo Y N, et al. Rice disease detection algorithm based on improved YOLOv8[J]. Jiangsu Journal of Agricultural Sciences, 2025, 41(3): 537-548 (in Chinese with English abstract).

[16] Duan L F, Yang W N, Bi K, et al. Fast discrimination and counting of filled/unfilled rice spikelets based on bi-modal imaging[J]. Computers and Electronics in Agriculture, 2011, 75(1): 196-203.

[17] Liu T, Ji C X, Wu W, et al. Measurement of rice filled grain percentage based on grain shadow features[J]. Agronomy Journal, 2016, 108(3): 1070-1078.

[18] Feng X, Wang Z Q, Zeng Z W, et al. Size measurement and filled/unfilled detection of rice grains using backlight image processing[J]. Frontiers in Plant Science, 2023, 14: 1213486.

[19] Zhou J, Lu X Y, Yang R, et al. Developing thermal infrared de-ghost and multi-level nested conglomerated segmentation algorithm for detection of rice seed setting rate[J]. Computers and Electronics in Agriculture, 2023, 207: 107725.

[20] Kong H H, Chen P. Mask R-CNN-based feature extraction and three-dimensional recognition of rice panicle CT images[J]. Plant Direct, 2021, 5(5): e00323.

[21] Guo Y X, Li S, Zhang Z G, et al. Automatic and accurate calculation of rice seed setting rate based on image segmentation and deep learning[J]. Frontiers in Plant Science, 2021, 12: 770916.

[22] Sun J, Ren Z W, Cui J L, et al. A high-throughput method for accurate extraction of intact rice panicle traits[J]. Plant Phenomics, 2024, 6: 213.

[23] 张续博, 张家瑜, 贲宗友, 等. 基于YOLOv5-Rf的糙米内部裂纹识别方法研究[J]. 南京农业大学学报,2024,47(4): 792-802.

[24] Zhang X B, Zhang J Y, Ben Z Y, et al. Research on internal crack recognition method of brown rice based on YOLOv5-RF[J]. Journal of Nanjing Agricultural University, 2024, 47(4): 792-802 (in Chinese with English abstract).

[25] Zhao Y A, Lv W Y, Xu S L, et al. DETRs beat YOLOs on real-time object detection[C]// 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 16-22, 2024, Seattle, WA, USA. IEEE, 2024: 16965-16974.

[26] Li H L, Li J, Wei H B, et al. Slim-neck by GSConv: a lightweight-design for real-time detector architectures[EB/OL]. 2022: arXiv: 2206.02424. https://arxiv.org/abs/2206.02424.

[27] Liu W Z, Lu H, Fu H T, et al. Learning to upsample by learning to sample[C]// 2023 IEEE/CVF International Conference on Computer Vision (ICCV). October 1-6, 2023, Paris, France. IEEE, 2023: 6004-6014.

[28] Rozemberczki B, Watson L, Bayer P, et al. The shapley value in machine learning[C]// Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence. July 23-29, 2022. Vienna, Austria. International Joint Conferences on Artificial Intelligence Organization, 2022: 5572-5579.

[29] Wang A, Chen H, Liu L H, et al. YOLOv10: real-time end-to-end object detection[EB/OL]. 2024: arXiv: 2405.14458. https://arxiv.org/abs/2405.14458.

基金资助

国家重点研发计划项目(2022YFD2300700)