Please wait a minute...
浙江大学学报(工学版)
浙江大学学报(工学版)  2026, Vol. 60 Issue (3): 604-613    DOI: 10.3785/j.issn.1008-973X.2026.03.016
计算机技术、控制工程     
轻量级改进RT-DETR的葡萄叶片病害检测算法
刘慧(),王防修*(),王意,黄淄博,苏晨
武汉轻工大学 数学与计算机学院，湖北 武汉 430048
Lightweight improved RT-DETR algorithm for grape leaf disease detection
Hui LIU(),Fangxiu WANG*(),Yi WANG,Zibo HUANG,Chen SU
School of Mathematics and Computer Science, Wuhan Polytechnic University, Wuhan 430048, China
 全文: PDF(6104 KB)   HTML
摘要：

针对葡萄叶片病害检测任务中存在的复杂背景干扰、小目标漏检及模型部署资源受限等问题，提出基于改进RT-DETR的轻量化检测算法SCGI-DETR. 引入高效轻量级的StarNet架构作为特征提取网络，减少模型的参数量和计算量，实现模型的轻量化. 设计CGSFR-FPN特征金字塔网络，通过空间特征重建和多尺度特征融合策略，增强模型对全局上下文信息的感知能力，提升复杂背景下多尺度病斑的定位精度. 构建Inner-PowerIoU v2损失函数，利用全局收敛加速与局部区域对齐机制，加速边界框回归，提高小目标检测性能. 实验结果表明，SCGI-DETR在葡萄叶片病害数据集上的精确率、召回率和mAP@0.5分别为91.6%、89.8%和93.4%，较原模型分别提升了2.6%、2.4%和2.3%，参数量与计算量分别减少了46.2%和64%. 该结果表明，改进算法在实现轻量化的同时，具备更优的检测性能，满足移动端和嵌入式设备的部署需求.
关键词： 葡萄叶片病害;  RT-DETR;  StarNet;  特征金字塔;  轻量化网络    
Abstract:

A lightweight detector SCGI-DETR was proposed based on an enhanced RT-DETR in order to address challenges in grape leaf disease detection—complex background, missed detection of small target, and resource-constrained deployment. The efficient StarNet backbone was employed to reduce parameter count and computational cost, enabling lightweight deployment. A feature pyramid CGSFR-FPN was designed. Spatial feature reconstruction was combined with multi-scale feature fusion in order to strengthen global context modeling and improve localization of multi-scale lesions in cluttered scenes. The Inner-PowerIoU v2 loss was constructed, which integrated global convergence acceleration and local region alignment in order to speed up bounding-box regression and enhance small-object detection performance. SCGI-DETR attained 91.6% precision, 89.8% recall and 93.4% mAP@0.5 on a grape leaf disease dataset, which improved 2.6, 2.4 and 2.3 percentage points over the baseline, and reduced parameters and computation by 46.2% and 64%, respectively. Results demonstrate that the improved algorithm achieves lightweight implementation while delivering superior detection performance, meeting deployment requirements for mobile and embedded devices.
Key words: grape leaf disease    RT-DETR    StarNet    feature pyramid    lightweight network
收稿日期: 2025-05-25 出版日期: 2026-02-04
:  TP 391  
基金资助: 湖北省高校优秀中青年科技创新团队资助项目(T2021009)；湖北省教育厅科技计划资助项目(D20211604).
通讯作者: 王防修     E-mail: 2283298757@qq.com;wfx323@whpu.edu.cn
作者简介: 刘慧（2002—），女，硕士生，从事目标检测、深度学习的农业工程应用研究. orcid.org/0009-0006-3184-7994. E-mail：2283298757@qq.com
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章　　
刘慧
王防修
王意
黄淄博
苏晨

引用本文:

刘慧,王防修,王意,黄淄博,苏晨. 轻量级改进RT-DETR的葡萄叶片病害检测算法[J]. 浙江大学学报(工学版), 2026, 60(3): 604-613.

Hui LIU,Fangxiu WANG,Yi WANG,Zibo HUANG,Chen SU. Lightweight improved RT-DETR algorithm for grape leaf disease detection. Journal of ZheJiang University (Engineering Science), 2026, 60(3): 604-613.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2026.03.016    或    https://www.zjujournals.com/eng/CN/Y2026/V60/I3/604

图 1  葡萄叶片病害的样本图
叶片类别$n$
训练集验证集测试集图像数量
黑麻疹病1 6584702482 376
黑腐病1 5664892082 263
叶枯病1 4664212292 116
健康1 7874722412 500
总数6 47718529269 255
表 1  葡萄病害数据集的图像数量
图 2  SCGI-DETR网络的结构图
图 3  StarNet网络的结构图
图 4  RCM模块的结构
图 5  PCE模块的结构
图 6  Inner-PowerIoU v2的结构图
StarNetCGSFR-FPNInner-PowerIoU v2P/%R/%mAP@0.5/%${N_{\text{P}}}$/106FLOPs/109
———89.087.491.119.957.0
√——90.388.492.212.031.8
—√—90.688.992.419.248.2
——√90.388.491.819.957.0
√√—90.789.592.710.720.5
√—√90.989.092.712.031.8
—√√89.988.691.819.248.2
√√√91.689.893.410.720.5
表 2  消融实验的对比结果
模型方法P/%R/%mAP@0.5/%${N_{\text{P}}}$/106$v$/(帧?s?1)FLOPs/109
RT-DETRResNet-18（原算法)89.087.491.119.967.257.0
RT-DETRFasterNet89.787.491.010.966.328.5
RT-DETREfficientViT88.987.590.810.743.527.2
RT-DETRRepViT89.887.191.213.357.636.3
RT-DETRMobileNetV489.787.891.711.370.739.5
RT-DETRStarNet90.388.492.212.072.231.8
表 3  不同轻量级主干网络的对比结果
损失函数P/%R/%mAP@0.5/%
GIoU（基线）89.087.491.1
EIoU88.887.290.8
SIoU89.587.491.0
PIoU90.287.391.3
PIoU v290.088.091.7
Inner-IoU90.287.791.4
Wise-IoU89.687.891.5
ShapeIoU90.087.891.3
Inner-PowerIoU v290.388.491.8
表 4  不同损失函数对SCGI-DETR的影响
$s$P/%R/%mAP@0.5/%
0.589.688.091.5
0.689.388.391.5
0.789.787.891.4
0.890.088.191.7
0.990.288.291.7
1.090.388.491.8
1.190.388.091.7
1.289.788.291.7
1.389.988.091.6
1.489.588.191.6
1.589.787.791.4
表 5  不同尺度因子对SCGI-DETR的影响
模型P/%R/%mAP@0.5/%${N_{\text{P}}}$/106$v$/(帧?s?1)FLOPs/109
SSD（2016）80.776.783.126.232.562.6
Faster R-CNN（2016）81.675.382.841.312.9212.8
EfficientDet（2020）83.778.684.533.413.4260.7
YOLOv7[29]（2022）89.888.792.636.482.8103.2
YOLOv8s（2023）91.089.593.011.1122.328.4
YOLOv9s[30]（2024）89.488.192.19.698.738.7
YOLOv10s[31]（2024）90.687.792.58.1130.224.5
YOLOv11s[32]（2024）91.288.693.19.4124.221.3
Deformable DETR（2020）89.188.091.039.910.8179.6
DINO（2022）89.487.992.346.77.3279.2
MS-DETR（2023）89.688.592.353.530.59117.1
RT-DETR（2023）89.087.491.119.967.257.0
SCGI-DETR（本文方法）91.689.893.410.766.720.5
表 6  不同模型的对比实验结果
图 7  小目标检测效果的对比
图 8  复杂背景下的检测效果对比
图 9  SCGI-DETR模型对叶片重点区域关注能力的可视化结果
1 岳喜申. 基于改进YOLOv5s的葡萄叶片病害识别方法研究 [D]. 阿拉尔: 塔里木大学, 2024.
YUE Xishen. A study on grape leaf disease identification method based on improved YOLOv5s [D]. Alar: Tarim University, 2024.
2 REN S, HE K, GIRSHICK R, 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
doi: 10.1109/TPAMI.2016.2577031
3 LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector [C]//14th European Conference on Computer Vision. Cham: Springer, 2016: 21–37.
4 REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection [C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 779–788.
5 姜晟, 曹亚芃, 刘梓伊, 等 基于改进Faster RCNN的茶叶叶部病害识别[J]. 华中农业大学学报, 2024, 43 (5): 41- 50
JIANG Sheng, CAO Yapeng, LIU Ziyi, et al Recognition of tea leaf disease based on improved Faster RCNN[J]. Journal of Huazhong Agricultural University, 2024, 43 (5): 41- 50
6 PAN P, SHAO M, HE P, et al Lightweight cotton diseases real-time detection model for resource-constrained devices in natural environments[J]. Frontiers in Plant Science, 2024, 15: 1383863
doi: 10.3389/fpls.2024.1383863
7 VASWANI A, SHAZEER N, PARMAR N, et al Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017, 30: 5998- 6008
8 ZHAO Y, LV W, XU S, et al. DETRs beat YOLOs on real-time object detection [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2024: 16965–16974.
9 HUANGFU Y, HUANG Z, YANG X, et al HHS-RT-DETR: a method for the detection of citrus greening disease[J]. Agronomy, 2024, 14 (12): 2900
doi: 10.3390/agronomy14122900
10 YANG H, DENG X, SHEN H, et al Disease detection and identification of rice leaf based on improved detection transformer[J]. Agriculture, 2023, 13 (7): 1361
doi: 10.3390/agriculture13071361
11 FU Z, YIN L, CUI C, et al A lightweight MHDI-DETR model for detecting grape leaf diseases[J]. Frontiers in Plant Science, 2024, 15: 1499911
doi: 10.3389/fpls.2024.1499911
12 MA X, DAI X, BAI Y, et al. Rewrite the stars [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2024: 5694–5703.
13 REZATOFIGHI H, TSOI N, GWAK J, et al. Generalized intersection over union: a metric and a loss for bounding box regression [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2020: 658–666.
14 NI Z, CHEN X, ZHAI Y, et al. Context-guided spatial feature reconstruction for efficient semantic segmentation [C]//European Conference on Computer Vision. Cham: Springer, 2024: 239–255.
15 LIU C, WANG K, LI Q, et al Powerful-IoU: more straightforward and faster bounding box regression loss with a nonmonotonic focusing mechanism[J]. Neural Networks, 2024, 170: 276- 284
doi: 10.1016/j.neunet.2023.11.041
16 ZHANG H, XU C, ZHANG S. Inner-IoU: more effective intersection over union loss with auxiliary bounding box [EB/OL]. (2023-11-06)[2025-03-16]. https://arxiv. org/pdf/2311.02877.
17 CHEN J, KAO S H, HE H, et al. Run, don’t walk: chasing higher FLOPS for faster neural networks [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 12021–12031.
18 LIU X, PENG H, ZHENG N, et al. EfficientViT: memory efficient vision transformer with cascaded group attention [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 14420–14430.
19 WANG A, CHEN H, LIN Z, et al. Rep ViT: revisiting mobile CNN from ViT perspective [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2024: 15909–15920.
20 QIN D, LEICHNER C, DELAKIS M, et al. MobileNetV4: universal models for the mobile ecosystem [C]// European Conference on Computer Vision. Cham: Springer, 2024: 78–96.
21 ZHANG Y F, REN W, ZHANG Z, et al Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 506: 146- 157
doi: 10.1016/j.neucom.2022.07.042
22 GEVORGYAN Z. SIoU loss: more powerful learning for bounding box regression [EB/OL]. (2022-05-25)[2025-03-16]. https://arxiv.org/abs/2205.12740.
23 TONG Z, CHEN Y, XU Z, et al. Wise-IoU: bounding box regression loss with dynamic focusing mechanism [EB/OL]. (2023-01-24)[2025-03-16]. https://arxiv.org/ abs/2301.10051.
24 ZHANG H, ZHANG S. Shape-IoU: more accurate metric considering bounding box shape and scale [EB/OL]. (2023-12-29)[2025-03-16]. https://arxiv.org/abs/ 2312.17663.
25 TAN M, PANG R, LE Q V. EfficientDet: scalable and efficient object detection [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 10778–10787.
26 ZHU X, SU W, LU L, et al. Deformable DETR: deformable transformers for end-to-end object detection [EB/OL]. (2020-10-08)[2025-03-16]. https://arxiv.org/ abs/2010.04159.
27 ZHANG H, LI F, LIU S, et al. DINO: DETR with improved denoising anchor boxes for end-to-end object detection [EB/OL]. (2022-03-07)[2025-03-16]. https://arxiv.org/abs/2203.03605.
28 ZHAO C, SUN Y, WANG W, et al. MS-DETR: efficient DETR training with mixed supervision [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2024: 17027–17036.
29 WANG C Y, BOCHKOVSKIY A, LIAO H M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 7464–7475.
30 WANG C Y, YEH I H, MARK LIAO H Y. YOLOv9: learning what you want to learn using programmable gradient information [C]//European Conference on Computer Vision. Cham: Springer, 2024: 1–21.
31 WANG A, CHEN H, LIU L, et al YOLOv10: real-time end-to-end object detection[J]. Advances in Neural Information Processing Systems, 2024, 37: 107984- 108011
32 KHANAM R, HUSSAIN M. YOLOv11: an overview of the key architectural enhancements [EB/OL]. (2024-10-23)[2025-03-16]. https://arxiv.org/abs/2410.17725.
[1] 梁耕良,韩曙光. 基于改进RT-DETR的牛仔面料疵点检测算法[J]. 浙江大学学报(工学版), 2025, 59(6): 1169-1178.
[2] 尹向雷,屈少鹏,解永芳,苏妮. 基于渐进特征融合及多尺度空洞注意力的遮挡鸟巢检测[J]. 浙江大学学报(工学版), 2025, 59(3): 535-545.
[3] 王安静,袁巨龙,朱勇建,陈聪,吴金津. 基于改进YOLOv8s的鼓形滚子表面缺陷检测算法[J]. 浙江大学学报(工学版), 2024, 58(2): 370-380.
[4] 杨淑琴,马玉浩,方铭宇,钱伟行,蔡洁萱,刘童. 基于实例分割的复杂环境车道线检测方法[J]. 浙江大学学报(工学版), 2022, 56(4): 809-815, 832.
[5] 谢誉,包梓群,张娜,吴彪,涂小妹,包晓安. 基于特征优化与深层次融合的目标检测算法[J]. 浙江大学学报(工学版), 2022, 56(12): 2403-2415.
[6] 李凯,林宇舜,吴晓琳,廖飞宇. 基于多尺度融合与注意力机制的小目标车辆检测[J]. 浙江大学学报(工学版), 2022, 56(11): 2241-2250.
[7] 董红召,方浩杰,张楠. 旋转框定位的多尺度再生物品目标检测算法[J]. 浙江大学学报(工学版), 2022, 56(1): 16-25.
[8] 徐利锋,黄海帆,丁维龙,范玉雷. 基于改进DenseNet的水果小目标检测[J]. 浙江大学学报(工学版), 2021, 55(2): 377-385.
[9] 林志洁,罗壮,赵磊,鲁东明. 特征金字塔多尺度全卷积目标检测算法[J]. 浙江大学学报(工学版), 2019, 53(3): 533-540.