CD-YOLO: 基于改进YOLOv11s的田间胡萝卜苗检测方法

欢迎您访问《智慧农业（中英文）》官方网站！English

Toggle navigation智慧农业

• 网站首页
• 期刊介绍
• 编委会
• 投稿指南
• 期刊服务
• 下载中心
• 联系我们

Smart Agriculture ›› 2026, Vol. 8 ›› Issue (2): 158-174.doi: 10.12133/j.smartag.SA202511008

• 信息处理与决策 •上一篇

CD-YOLO：基于改进YOLOv11s的田间胡萝卜苗检测方法

刘浩然 1,2, 王宇 1, 赵学观 2,4, 吴华瑞 3, 付豪 2,4, 庞树杰 5, 翟长远 2,4()

1. 1. 辽宁科技大学机械工程与自动化学院，辽宁 鞍山 114051，中国

2. 北京市农林科学院智能装备技术研究中心，北京 100097，中国

3. 北京市农林科学院信息技术研究中心，北京 100097，中国

4. 国家农业智能装备工程技术研究中心，北京 100097，中国

5. 中国农业机械化科学研究院集团有限公司，北京 100083，中国

• 收稿日期:2025-11-09 出版日期:2026-03-30
• 基金项目:北京市农林科学院院创新能力建设专项(KJCX20230409); 国家自然科学基金(32201647); 改革与发展项目(GGFZ20250205)
• 作者简介:刘浩然，硕士研究生，研究方向为智能化施药装备研发。E-mail：[email protected]
• 通信作者:翟长远，博士，研究员，研究方向为智能化施药装备研发。E-mail：[email protected]

CD-YOLO: A Method for Detecting Carrot Seedlings in Field Based on Improved YOLOv11s

LIU Haoran 1,2, WANG Yu 1, ZHAO Xueguan 2,4, WU Huarui 3, FU Hao 2,4, PANG Shujie 5, ZHAI Changyuan 2,4()

1. 1. School of Mechanical Engineering and Automation, University of Science and Technology Liaoning, Anshan 114051, China

2. Intelligent Equipment Research Center, Beijing Academy of Agriculture and Forestry Sciences, Beijing 100097, China

3. Information Technology Research Center, Beijing Academy of Agriculture and Forestry Sciences, Beijing 100097, China

4. National Engineering Research Center of Intelligent Equipment for Agriculture (NERCIEA), Beijing 100097, China

5. Chinese Academy of Agricultural Mechanization Sciences Group Co. , Ltd. , Beijing 100083, China

• Received:2025-11-09 Online:2026-03-30
• Foundation items:Beijing Academy of Agriculture and Forestry Sciences Innovation Capacity Building Project(KJCX20230409); National Natural Science Foundation of China(32201647); Reform and Development Project(GGFZ20250205)
• About author:LIU Haoran, E-mail: [email protected]
• Corresponding author:ZHAI Changyuan, E-mail: [email protected]

在线阅读 38

知网下载

本地下载 42

## 可视化 0

摘要/Abstract

摘要：

【目的/意义】 枝叶遮挡是田间胡萝卜苗精准识别的主要挑战，对此开发一种轻量、高效的检测模型，以实现田间遮挡条件下胡萝卜苗的精准鲁棒识别。 【方法】 提出了一种基于改进YOLOv11s（You Only Look Once）的轻量化检测方法CD-YOLO（Carrot Detection-YOLO）。首先，为降低模型复杂度，将主干网络中的部分标准卷积替换为深度可分离卷积，以减少计算量与参数量；其次，在C3k2模块中引入高效多尺度注意力机制（Efficient Multi-Scale Attention, EMA），构建C3k2_EMA模块，增强模型对关键特征的动态感知能力，从而抑制背景与遮挡噪声；最后，采用动态检测头，强化模型对不同尺度特征的融合与感知，进一步提升检测鲁棒性。 【结果和讨论】 结果表明：CD-YOLO模型的计算量和模型大小相较于YOLOv11s分别降低了6.2 GFLOPs和4.8 MB，单张图像推理速度提升了4.7 ms，准确率、召回率与平均精度分别提高了3.0、1.5和2.4个百分点；在应对枝叶遮挡方面，CD-YOLO的漏识率为13.4%，较YOLOv11s下降了5.7个百分点；将模型部署于边缘算力设备，在随机选取的胡萝卜苗图像上进行测试，漏识率和误识率分别为5.1%和2.7%。 【结论】 CD-YOLO模型在保持轻量化的同时，有效提升了田间胡萝卜苗的检测精度与速度，特别是在应对枝叶遮挡方面表现优异。该研究可为胡萝卜自动化育苗过程中的出苗率统计与精准管理提供可靠的技术支持。

关键词:胡萝卜苗,遮挡,目标检测,YOLOv11s,轻量化

Abstract:

[Objective] In field environments under natural conditions, leaf occlusion and mutual plant shading pose significant challenges to the accurate identification of carrot seedlings. Furthermore, practical agricultural applications often rely on edge devices with limited computational power, necessitating a detection model that combines lightweight design, high accuracy, and robust anti-occlusion capability. The purpose of this research is to develop a robust recognition method for carrot seedlings suitable for complex field conditions, thereby enhancing the accuracy and efficiency of seedling emergence statistics in automated seedling raising processes and providing reliable technical support for precise farm management. [Methods] The CD-YOLO (Carrot Detection-You Only Look Once), a lightweight detection model was proposed based on an improved YOLOv11s. First, to reduce model complexity, several standard convolutions in the backbone network were replaced with depthwise separable convolutions (DWConv), thereby decreasing floating-point operations (FLOPs) and the number of parameters, establishing a lightweight foundation for edge deployment. Secondly, the efficient multi scale attention (EMA) mechanism was embedded into the critical feature extraction module C3k2, constructing a C3k2_EMA module. This module enhanced dynamic perception of local key features and reconstructed cross-scale contextual dependencies broken by occlusion through its parallel multi-branch structure, effectively suppressing background and occlusion noise. Finally, the DynamicHead detection head was introduced. Leveraging its scale-aware and spatial-aware mechanisms, it achieved a dynamic fusion of multi-level features and adaptive weight adjustment, further improving the model's decision-making robustness in complex scenes. To comprehensively evaluate model performance, a carrot seedling dataset covering various field scenarios was independently constructed. Through offline data augmentation, the original 1 274 images were expanded to 4 796, which were then split into training, validation, and test sets in an 8:1:1 ratio. Meanwhile, to systematically quantify the model's anti-occlusion performance, an occlusion severity assessment criterion based on the overlapping area of bounding boxes was proposed. Targets were categorized into three occlusion levels: mild, moderate, and severe. Based on this, a dedicated "Occlusion Test Subset" was separated from the main test set, providing an objective and reproducible benchmark for evaluating the model's anti-occlusion capability. [Results and Discussions] Experimental results on the custom dataset demonstrated that CD-YOLO comprehensively improved detection performance while maintaining its lightweight characteristics. Compared to the baseline model YOLOv11s, CD-YOLO reduced computational load by 6.2 GFLOPs (a 28.8% decrease), decreased model size by 4.8 MB (a 25.0% reduction), improved single-image inference speed by 4.7 ms, reaching 9.6 ms. Concurrently, precision, recall, and mean average precision (mAP 0.5) increased by 3.0, 1.5, and 2.4 percentage points, respectively, ultimately reaching 81.2%, 76.4%, and 84.0%. In comparisons with other lightweight backbone networks like MobileNetv3 and ShuffleNetv2, CD-YOLO consistently outperformed them on the accuracy-speed comprehensive metric, validating the effectiveness of its improvement strategies. In occlusion performance tests, the missed detection rate of CD-YOLO on the occlusion test subset was 13.4%, a 5.7 percentage points decrease compared to YOLOv11s. Its mAP 0.5 on the occlusion subset reached 80.6%, a 5.1 percentage points improvement over the baseline, whereas the improvement on the regular subset was 1.8 percentage points, proving the model's enhanced efficacy in occlusion scenarios. After deploying the model on an NVIDIA Jetson Orin NX edge device and accelerating it with TensorRT, the inference frame rate increased to 32.5 f/s. On random test images, CD-YOLO achieved missed detection and false detection rates of 5.1% and 2.7%, respectively, representing decreases of 7.7% and 2.6% compared to YOLOv11s, demonstrating promising practical application potential. Ablation studies and feature map visualizations further indicated that DWConv, C3k2_EMA, and DynamicHead formed a synergistic optimization loop: DWConv achieved computational compression, freeing up computational budget for subsequent modules; C3k2_EMA enhanced local perception and contextual reconstruction of occluded targets during the feature extraction stage; and DynamicHead performed dynamic fusion of multi-scale features at the decision-making end. Together, they ensured high-precision detection of incomplete targets under limited computational resources. [Conclusions] Through the synergistic design of "lightweighting, feature enhancement, and dynamic fusion", the CD-YOLO model achieved an excellent balance between computational efficiency, detection accuracy, and anti-occlusion capability. The model not only significantly reduced reliance on the computational power of edge devices but also effectively improved robustness and adaptability in complex field environments through structured attention and dynamic fusion mechanisms.

Key words:carrot seedlings,occlusion,object detection,YOLOv11s,lightweight

中图分类号:

• TP391.4

• S24

引用本文

刘浩然, 王宇, 赵学观, 吴华瑞, 付豪, 庞树杰, 翟长远. CD-YOLO：基于改进YOLOv11s的田间胡萝卜苗检测方法[J]. 智慧农业(中英文), 2026, 8(2): 158-174.

LIU Haoran, WANG Yu, ZHAO Xueguan, WU Huarui, FU Hao, PANG Shujie, ZHAI Changyuan. CD-YOLO: A Method for Detecting Carrot Seedlings in Field Based on Improved YOLOv11s[J]. Smart Agriculture, 2026, 8(2): 158-174.

使用本文

/[推荐](javascript:void(null))
导出引用管理器EndNote|Ris|BibTeX

链接本文:https://www.smartag.net.cn/CN/10.12133/j.smartag.SA202511008

https://www.smartag.net.cn/CN/Y2026/V8/I2/158

[图/表 20](javascript:;)

图1

胡萝卜苗采集图像实例

图1

图2

胡萝卜苗遮挡程度量化图

图2

表1

胡萝卜苗数据集分布表

类别	图像数量/张	标签数量/张
训练集	3 837	43 880
验证集	480	6 156
测试集	479	6 867

表1

图3

CD-YOLO结构图

图3

图4

EMA模块

图4

图5

C3k2_EMA模块

图5

表2

CD-YOLO模型训练参数

训练参数	值
训练周期	100
初始学习率	0.001
优化器	Adam
每批次图像数量	24
动量	0.937
优化器权重衰减系数	0.000 5
输入尺寸	640×640

表2

表3

胡萝卜苗检测研究消融试验结果

试验编号	改进方式	P/%	R/%	mAP 0.5/%	计算量/GFLOPs	模型大小/MB	单张图片处理时间/ms
DWConv	C3k2_EMA	DynamicHead
1	×	×	×	78.2	74.9	81.6	21.5
2	√	×	×	79.8	73.6	81.4	14.8
3	×	√	×	78.7	77.1	82.9	21.3
4	×	×	√	79.9	76.7	83.7	21.5
5	√	√	×	79.7	74.5	81.7	14.7
6	√	×	√	79.0	74.9	81.8	15.0
7	×	√	√	80.2	75.1	83.2	21.6
8	√	√	√	81.2	76.4	84.0	15.3

表3

图6

C3k2与C3k2_EMA模块特征图对比

图6

图7

不同模型在胡萝卜苗测试集上的效果对比

图7

图8

YOLOv11s与CD-YOLO在遮挡条件下胡萝卜苗检测效果对比 注：图中蓝色框为检测框，红色框为漏检目标。

图8

表4

不同轻量化主干网络检测胡萝卜苗效果对比

主干网络	P/%	R/%	mAP 0.5/%	计算量/GFLOPs	参数量	模型大小/MB	单张图片处理时间/ms
MobileNetv3	76.2	73.0	79.9	13.2	7.21×10 6	16.9	14.6
ShuffleNetv2	73.0	72.3	77.8	10.4	5.34×10 6	11.0	12.2
EfficientVit	76.0	74.5	79.7	14.6	7.39×10 6	15.7	11.4
CD-YOLO	81.2	76.4	84.0	15.3	7.06×10 6	14.4	9.6

表4

表5

胡萝卜苗检测研究不同模型对比效果

模型	P/%	R/%	mAP 0.5/%	计算量/GFLOPs	模型大小/MB	单张图片处理时间/ms
SSD	63.5	81.4	72.2	63.5	90.5	36.6
Faster R-CNN	69.1	84.5	77.7	142.6	108.1	113.3
YOLOv5s	80.2	73.9	80.5	16.2	15.1	14.2
YOLOv8s	79.1	74.8	81.6	49.1	22.5	9.8
YOLOv11s	78.2	74.9	81.6	21.5	19.2	14.3
YOLOv8s-P2	79.3	74.4	81.5	55.3	24.8	10.7
DWG-YOLOv8	78.2	73.2	82.0	17.4	14.7	10.5
HAD-YOLO	77.9	74.3	81.1	35.4	12.5	19.3
CD-YOLO	81.2	76.4	84.0	15.3	14.4	9.6

表5

表6

YOLOv11s与CD-YOLO在不同测试子集上效果对比

测试子集	模型	图像数量	P/%	R/%	mAP 0.5/%
常规子集	YOLOv11s	357	78.7	76.3	77.6
CD-YOLO	357	79.8	78.1	79.4
遮挡子集	YOLOv11s	122	75.9	71.1	75.5
CD-YOLO	122	79.6	75.1	80.6
整体测试集	YOLOv11s	479	78.9	74.7	76.9
CD-YOLO	479	79.7	76.2	79.9

表6

表7

YOLOv11s与CD-YOLO遮挡条件下胡萝卜苗检测效果对比

模型	目标总数/个	正确识别数/个	漏识数/个	识别率/%	漏识率/%
YOLOv11s	1 652	1 337	315	80.9	19.1
CD-YOLO	1 652	1 431	221	86.6	13.4

表7

图9

YOLOv11s与CD-YOLO模型热力图可视化对比

图9

表8

YOLOv11s与CD-YOLO模型在不同设备上检测帧率对比 ( 帧/s)

模型	工作站	边缘算力设备	TensorRT
YOLOv11s	89.6	10.5	29.7
CD-YOLO	104.2	12.9	32.5

表8

表9

YOLOv1