The apperance quality of Pu′er Dragon Ball tea plays a decisive role in its market value; however, conventional inspection approaches fail to simultaneously satisfy the demands of real-time efficiency, accuracy, and edge-level deployment. In response, we propose SHM-YOLO, a lightweight object detection framework. Extending YOLOv11, the model employs ShuffleNetV2 (denoted as S in SHM) as the backbone, integrating point wise group convolution with channel shuffling to minimize computational cost. Through the integration of a hierarchical scale feature pyramid network (HS-FPN, denoted as H in SHM) that combines channel attention with dimensional matching, the model strengthens the effectiveness of multi-scale feature fusion. At the same time, the multi-scale attention block (MAB, denoted as M in SHM) is utilized to optimize the C3K2 structure, enabling more effective image detail extraction. To improve bounding-box regression, the model combines Inner-IoU with SIoU loss, which expedites convergence and augments localization precision. Experimental validation on a self-developed dataset for Pu′er Dragon Ball tea appearance quality confirms that SHM-YOLO reaches 97.2% mAP@50, 92.7% precision (P), 93.6% recall (R), and 303 fps, with merely 0.969×10⁶parameters and 2.3 MB storage consumption. Compared to YOLOv11n, the model achieves higher accuracy while markedly decreasing floating-point computation (by 62.5%) and memory consumption (by 47.6%), highlighting its excellent lightweight characteristics and strong suitability for industrial deployment.
除此之外本研究还使用了梯度加权类激活图(gradient-weighted class activation mapping)来可视化,如图11(c)所示。从类激活图11(c)中可以观察到,SHM-YOLO比YOLOv11的轮廓的红色区域更深更大。这表明普洱龙珠茶的特征更加清晰,使得网络能够专注于更有价值的特征区域。综上所述,本研究通过对图像采集预处理、神经网络架构和评价指标优化等方面实施一系列改进策略,提出一套普洱龙珠茶分级检测的方案。这使得模型在茶叶形态相似、色泽相近等复杂情况下仍能有效运作。该模型不仅能精准地对普洱龙珠茶进行分级,还显著提升了检测效率,在普通的边缘生产设备上也能展现出良好性能。本研究的改进方法对其他茶叶品种的分级以及农产品品质检测领域具有较高的参考价值。不过,本研究也存在一定局限,例如模型对于部分特征极为相似的普洱龙珠茶的识别准确率,还有提升空间。在未来,将针对这些问题展开深入研究。
ZANGG H, ZHANGL Y.Application and development prospects of intelligent agricultural machinery and equipment in modern agricultural production[J].China Agricultural Machinery Equipment,2025(3):137-139.
[3]
ZHANGG F.Analysis of the application advantages and technical characteristics of agricultural machinery automation technology[J].Forum on Research and Innovation Management,2024,2(5):4520.
GAOY C, XUC, LINQ,et al.Tea production equipment automation and intelligent technology research progress and prospects[J].Transactions of the Chinese Society for Agricultural Machinery,2024,55(7):1-14.
CHANGY M, DONGY H, LINW Y,et al.Evaluation of high-quality development level of the tea industry:A case study of major tea-producing areas in Fujian Province[J].China Agricultural Resources and Regional Planning,2025:1-17.
BAOJ G, CHENY Y, DONGY H.Mechanisms of rural revitalization through Pu’er tea industry development:A case study of Yiwu,Yunnan[J].Journal of Natural Resources,2023,38(8):1941-1954.
General Administration of Quality Supervision,Inspection and Quarantine of the People’s Republic of China,Standardization Administration of the People’s Republic of China. Technique specification for producing and manufacturing of brick tea:GB/T 24615—2009 [S].Beijing:Standards Press of China,2009.
LYU Q, LING, JIANGJ,et al.Detecting green Citrus fruit in natural scenes using improved YOLOv5s model[J].Transactions of the Chinese Society of Agricultural Engineering,2024,40(18):147-154.
LIY, LIUM L, HEZ F,et al.Detecting Citrus fruit maturity using improved YOLOv8s[J].Transactions of the Chinese Society of Agricultural Engineering,2024,40(24):157-164.
[20]
ZHANGY F, LIL, CHUNC P,et al.Multi-scale feature adaptive fusion model for real-time detection in complex Citrus orchard environments[J].Computers and Electronics in Agriculture,2024,219:108836.
[21]
李力扬.基于改进YOLOv8的草莓识别算法[J].南方农机,2025,56(5):16-20.
[22]
LIL Y.Strawberry recognition algorithm based on an improved YOLOv8 model[J].Southern Agricultural Machinery,2025,56(5):16-20.
GUOJ T,LYU F, ZHANGH T,et al.Strawberry target detection method based on transfer learning and lightweight YOLOv5s[J].Journal of Chinese Agricultural Mechanization,2025,46(3):253-260
[25]
LINZ X, LIUW, WANGS Y,et al.Strawberry harvest date prediction using multi-feature fusion deep learning in plant factory[J].Computers and Electronics in Agriculture,2025,234:110174.
TANGY S, XUR H, WANGS J.Research progress in the application of machine vision in food nondestructive detection[J].Journal of Chinese Institute of Food Science and Technology,2024,24(12):13-27.
HEJ B, LIUC Z, GAOS.Development status and application prospects of machine vision technology in agriculture[J].Agricultural Development and Equipment,2025(1):50-52.
[30]
DHANUSHG, KHATRIN, KUMARS,et al.A comprehensive review of machine vision systems and artificial intelligence algorithms for the detection and harvesting of agricultural produce[J].Scientific African,2023,21:e01798.
[31]
MAN N, ZHANGX Y, ZHENGH T,et al.ShuffleNet V2:Practical guidelines for efficient CNN architecture design[M]// Computer Vision – ECCV 2018.Cham:Springer International Publishing,2018:122-138.
[32]
CHENY F, ZHANGC Y, CHENB,et al.Accurate leukocyte detection based on deformable-DETR and multi-level feature fusion for aiding diagnosis of blood diseases[J].Computers in Biology and Medicine,2024,170:107917.
[33]
WANGL, SHENJ, TANGE,et al.Multi-scale attention network for image super-resolution[J].Journal of Visual Communication and Image Representation,2021,80:103300.
[34]
GEVORGYANZ.SIoU loss:More powerful learning for bounding box regression[J].arXiv,2205:12740.
[35]
RENS Q, HEK M, GIRSHICKR,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2017,39(6):1137-1149.
[36]
HEK, GKIOXARIG, DOLLÁRP,et al.Mask R-CNN[C]//2017 IEEE International Conference on Computer Vision (ICCV).October 22-29,2017,Venice,Italy.IEEE,2017:2980-2988.
[37]
KHANAMR, HUSSAINM.YOLOv11:An overview of the key architectural enhancements[J].arXiv,2024:17725.
[38]
ALIM L, ZHANGZ.The YOLO framework:A comprehensive review of evolution,applications,and benchmarks in object detection[J].Computers,2024,13(12):336.
[39]
KIOBYAT, ZHOUJ F, MAISELIB.A multi-scale semantically enriched feature pyramid network with enhanced focal loss for small-object detection[J].Knowledge-Based Systems,2025,310:113003.
[40]
LIZ Q, SUY X, ZHANGY H,et al.Remote sensing image classification method based on improved ShuffleNet convolutional neural network[J].Intelligent Data Analysis,2024,28(2):397-414.
[41]
ZHANGH, XUC, ZHANGS.Inner-IoU:More effective intersection over union loss with auxiliary bounding box[EB/OL].arXiv:2311.07916,2023-11-14[2025-09-05].
[42]
TRANQ, MITRAJ, NGUYENN.Learning model combining convolutional deep neural network with a self-attention mechanism for AC optimal power flow[J].Electric Power Systems Research,2024,231:110327.
[43]
HUANGY, FUG, RENW C,et al.Low-light images enhancement via a dense transformer network[J].Digital Signal Processing,2024,148:104467.
[44]
WANGY F, XUX S, WANGZ,et al.ShuffleNet-Triplet:A lightweight RE-identification network for dairy cows in natural scenes[J].Computers and Electronics in Agriculture,2023,205:107632.
XIAOT Y, AIT H, YUH F,et al.A point cluster simplification approach of graph convolutional neural network for map generalization[J].Acta Geodaetica et Cartographica Sinica,2024,53(1):158-172.
LIJ Y, YANZ G, YANK D,et al.Detection algorithm of ground target based on attention and channel rearrangement for UAV[J].Journal of Ordnance Equipment Engineering,2024,45(3):306-313.
[49]
WOOH J, SIMJ W, KIME T.Lightweight super-resolution network with information distillation and recursive methods[C]//2022 IEEE International Conference on Consumer Electronics (ICCE).January 7-9,2022,Las Vegas,NV,USA.IEEE,2022:1-2.
HUIZ, GAOX B, YANGY C,et al.Lightweight image super-resolution with information multi-distillation network[C]//Proceedings of the 27th ACM International Conference on Multimedia.October 21-25,2019,Nice France.ACM,2019:2024-2032.
[54]
WANGW H, XIEE Z, LIX,et al.Pyramid vision transformer:A versatile backbone for dense prediction without convolutions[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV).October 10-17,2021,Montreal,QC,Canada.IEEE,2021:548-558.
[55]
LIQ X, MAW J, LIH,et al.Cotton-YOLO:Improved YOLOv7 for rapid detection of foreign fibers in seed cotton[J].Computers and Electronics in Agriculture,2024,219:108752.
[56]
HEW L, WUF, SNYDERL U,et al.Improved two-stage deep learning algorithm and lightweight YOLOv5n for classifying cottonseed damage[J].Computers and Electronics in Agriculture,2025,232:110042.