基于改进YOLOv10和LAMP通道剪枝的串番茄成熟度检测算法研发

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): 133-146.doi: 10.12133/j.smartag.SA202507045

• 信息处理与决策 •上一篇

基于改进YOLOv10和LAMP通道剪枝的串番茄成熟度检测算法研发

赵丽成 1,2, 卢鑫羽 2, 吴茜 2, 任妮 2, 周玲莉 2, 程雅雯 2, 胡安琦 2, 戚超 2()

1. 1. 淮阴工学院化学工程学院，江苏 淮安 223003，中国

2. 江苏省农业科学院农业信息研究所，江苏 南京 210014，中国

• 收稿日期:2025-07-30 出版日期:2026-03-30
• 基金项目:国家自然科学基金(32201664); 江苏省农业科技自主创新资金（CX（24）1021）(CX(24)1021); 农机研发制造推广应用一体化试点专项(JSYTH08)
• 作者简介:赵丽成，硕士研究生，研究方向为作物表型算法。E-mail：[email protected]
• 通信作者:戚 超，博士，助理研究员，研究方向为图像处理与机器视觉。E-mail：[email protected]

An Improved YOLOv10-Based Tomato Ripeness Detection Algorithm with LAMP Channel Pruning

ZHAO Licheng 1,2, LU Xinyu 2, WU Qian 2, REN Ni 2, ZHOU Lingli 2, CHENG Yawen 2, HU Anqi 2, QI Chao 2()

1. 1. School of Chemical Engineering, Huaiyin Institute of Technology, Huai'an 223003, China

2. Institute of Agricultural Information, Jiangsu Academy of Agricultural Sciences, Nanjing 210014, China

• Received:2025-07-30 Online:2026-03-30
• Foundation items:National Natural Science Foundation of China(32201664); Jiangsu Provincial Agricultural Science and Technology Independent Innovation Fund(CX(24)1021); Special Pilot Program for the Integrated R&D, Manufacturing, Promotion and Application of Agricultural Machinery(JSYTH08)
• About author:biography:ZHAO Licheng, E-mail: [email protected]
• Corresponding author:QI Chao, E-mail: [email protected]

在线阅读 33

知网下载

本地下载 63

## 可视化 0

摘要/Abstract

摘要：

【目的/意义】 由于串番茄果实密集重叠、温室光照复杂多变及果实颜色连续渐变等因素，检测模型在对成熟度进行精准快速识别时面临挑战。因此，提出了一种基于改进YOLOv10（You Only Look Once version 10）的轻量级串番茄成熟度目标检测模型LampCT-YOLO （Cluster Tomato YOLO with Layer-wise Adaptive Mask Pruning），以提高检测精度、推理速度和鲁棒性。 【方法】 以YOLOv10为基线模型，在主干网络引入SegNeXt（Segmentation Next）注意力机制，通过自适应调整注意力权重，增强模型对3类不同成熟度串番茄关键区域（如边界、颜色深浅）的特征提取能力；在训练完成的模型基础上，通过基于梯度的全局通道重要性方法LAMP（Layer-wise Adaptive Mask Pruning）通道剪枝，显著压缩模型体积和计算复杂度，同时有效保持模型对串番茄3分类成熟度的高检测性能。 【结果和讨论】 在NVIDIA A100显卡环境下，LampCT-YOLO模型对串番茄红熟早期、中期、晚期的平均精度均值（Mean Average Precision, mAP）分别为84.6%、89.5%、88.4%，相比YOLOv10分别提升了5.5、7.7和0.9个百分点，对串番茄3个成熟度类别的mAP 50 为87.6%，相比YOLOv10提升了4.7个百分点。应用LAMP通道剪枝技术后，模型的参数量和计算量分别减少63.07%和50.06%，推理速度提升23.1%。部署至自主研发的果蔬巡检机器人后，在边缘设备NVIDIA Jetson AGX Orin上达到9.1帧/s的推理速度，实现了检测精度与实时性的有效平衡。 【结论】 研究结果可为基于巡检机器人的串番茄成熟度精准快速识别提供坚实的技术支撑。

关键词:串番茄,成熟度检测,注意力机制,通道剪枝,果蔬巡检机器人,YOLOv10

Abstract:

[Objective] As a major crop in protected horticulture, cluster tomatoes grow in clusters with dense overlapping fruits. In greenhouse environments, light conditions are complex and variable, and the fruit color transitions continuously from green to red across different ripening stages, showing continuous gradation characteristics. These factors result in the low efficiency and strong subjectivity of traditional manual recognition methods. Meanwhile, deep learning-based detection models often suffer from decreased detection accuracy, large localization errors, and slow inference speed when facing complex backgrounds and color interference, making it difficult to meet the dual requirements of real-time performance and high precision in practical applications. Therefore, to meet the practical application requirements of high accuracy, high real-time performance, and strong robustness for cluster tomato ripeness detection, this paper proposes a lightweight target detection model for cluster tomato ripeness, namely LampCT-YOLO (Cluster Tomato YOLO with LAMP pruning), which is based on improved YOLOv10. Through structural optimization and lightweight transformation of the baseline model, the detection accuracy, inference speed, and robustness are effectively improved, providing a novel technical solution for cluster tomato ripeness detection. [Methods] Taking YOLOv10 as the baseline model, first, the issue of insufficient feature extraction capability in complex scenarios was addressed by introducing the SegNeXt attention mechanism into the backbone network. By adaptively adjusting attention weights and calculating the correlation matrix between different feature channels, the mechanism automatically identified color channels strongly associated with the three ripeness levels of cluster tomatoes and assigned them higher attention weights, while suppressing feature responses from irrelevant background channels such as greenhouse frames, soil, and irrigation pipes. To achieve lightweight deployment of the model and meet the real-time detection requirements of edge devices, a gradient-based global channel importance method—LAMP channel pruning technology—was introduced after model training. The core principle of this technology was to evaluate the contribution of each channel to the model's detection performance by calculating the gradient magnitude of channels in each network layer, thereby eliminating redundant channels. This significantly reduced the model size and computational complexity while effectively maintaining the model's high detection performance for the three-category ripeness classification of cluster tomatoes. [Results and Discussions] Experiments showed that under the environment of NVIDIA A100 graphics card, for 240 cluster tomato images in the test set, the LampCT-YOLO model exhibited excellent detection performance. The mean average precision at 50 intersection over union (mAP50) for the early ripe, mid-ripe, and late ripe stages of cluster tomatoes was 84.6%, 89.5%, and 88.4%, respectively, which represented increases of 5.5, 7.7, and 0.9 percentage points compared with YOLOv10. The average mAP50 for the three ripeness categories of cluster tomatoes reached 87.6%, a 4.7 percentage points improvement over YOLOv10, demonstrating outstanding performance in both detection accuracy and stability. In addition, the model was found to maintain high recognition accuracy when facing variations in light intensity, fruit occlusion ratio, and background complexity, indicating good robustness and environmental adaptability. Regarding the lightweight effect, after applying the LAMP channel pruning technology, the number of model parameters and computational complexity were reduced by 63.07% and 50.06%, respectively, while the inference speed was improved by 23.1%. This effectively met the requirements of edge computing devices for real-time detection and low power consumption, alleviating the trade-off between model accuracy and inference speed. To verify the practical application value of the LampCT-YOLO model, the model was deployed on a self-developed fruit and vegetable inspection robot, which conducted field tests on 456 clusters of tomatoes in a real greenhouse environment. The results showed that the inspection robot successfully identified 78, 61, and 248 clusters of early ripe, mid-ripe, and late ripe cluster tomatoes, respectively, with detection accuracies of 84.8%, 87.1%, and 84.4%, and an average accuracy of 85.4%. Meanwhile, there were 5, 7, and 10 false detections, as well as 9, 2, and 36 missed detections for the early ripe, mid-ripe, and late ripe stages respectively, which to a certain extent reflected the practical application potential of the model. [Conclusions] The optimized LampCT-YOLO model not only significantly improves the recognition accuracy of cluster tomatoes at different ripening stages but also greatly reduces the model complexity, successfully achieving efficient deployment in resource-constrained scenarios. This model effectively balances the dual requirements of detection accuracy and real-time performance for inspection robots, and further constructs a reusable technical framework for the ripeness detection of protected horticultural fruits and vegetables. It provides strong support for the transformation of protected agriculture from labor-intensive to technology-intensive, and injects key innovative impetus into the large-scale and diversified implementation of smart agriculture.

Key words:cluster tomato,ripeness detection,attention mechanism,channel pruning,fruit and vegetable inspection robot,YOLOv10

中图分类号:

• TP391.41

• S641.2

• S626

引用本文

赵丽成, 卢鑫羽, 吴茜, 任妮, 周玲莉, 程雅雯, 胡安琦, 戚超. 基于改进YOLOv10和LAMP通道剪枝的串番茄成熟度检测算法研发[J]. 智慧农业(中英文), 2026, 8(2): 133-146.

ZHAO Licheng, LU Xinyu, WU Qian, REN Ni, ZHOU Lingli, CHENG Yawen, HU Anqi, QI Chao. An Improved YOLOv10-Based Tomato Ripeness Detection Algorithm with LAMP Channel Pruning[J]. Smart Agriculture, 2026, 8(2): 133-146.

使用本文

[收藏文章](javascript:;) /[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.SA202507045

https://www.smartag.net.cn/CN/Y2026/V8/I2/133

[图/表 18](javascript:;)

图1

移动巡检机器人现场图

图1

图2

串番茄不同成熟度示例图

图2

表1

串番茄三类成熟度标注样本数量

成熟时期	训练集	验证集	测试集	总计（占比/%）
红熟早期	93	841	148	1 082（32.07）
红熟中期	104	934	166	1 204（35.07）
红熟晚期	107	831	149	1 087（32.86）

表1

图3

串番茄图像数据增强示例图

图3

图4

LampCT-YOLO模型结构图

图4

图5

SegNeXt注意力机制网络结构图

图5

图6

LAMP通道剪枝示意图

图6

表2

串番茄成熟度检测的SegNeXt注意力机制消融试验

模型	红熟早期/%	红熟中期/%	红熟晚期/%	mAP 50/%	FPS/（帧/s）	计算量/GFLOPs	参数量/MB	权重文件大小/MB
YOLOv10	79.1	81.8	87.5	82.8	37.37	106.700	45.811	107.7
YOLOv10+SegNeXt	84.6（+5.5）	89.5（+7.7）	88.4（+0.9）	87.5（+4.7）	35.8（-1.57）	228.765	53.522	107.6（-0.1）

表2

表3

Lamp模型剪枝消融试验

模型	红熟早期/%	红熟中期/%	红熟晚期/%	mAP 50/%	FPS/（帧/s）	计算量/GFLOPs	参数量/MB	权重文件大小/MB
YOLOv10+SegNeXt	84.6	89.5	88.4	87.5	35.8	228.7	53.5	107.6
YOLOv10+SegNeXt +DWConv	75.4 （-9.2）	75.0 （-14.5）	82.2 （-6.2）	77.5 （-10.0）	37.7 （+1.9）	115.2 （-113.5）	21.1 （-32.4）	52.2 （-55.5）
YOLOv10+SegNeXt +Lamp	81.4 （-3.2）	84.3 （-5.2）	91.9 （+3.5）	85.9 （-1.6）	66.9 （+23.1）	114.2 （-114.5）	19.7 （-33.7）	40.0 （-67.7）

表3

图7

LampCT-YOLO模型优化前后注意力可视化对比

图7

图8

LampCT-YOLO模型各层剪枝前后参数对比图

图8

图9

LampCT-YOLO模型置信区间与分布小提琴图

图9

表4

增加注意力机制后模型与主流目标检测模型对比试验

模型名称	mAP 50/%	FPS/（帧/s）	计算量/GFLOPs	模型参数量/MB	权重文件大小/MB
SSD	77.3	43.22	37.8	21.472	91.6
Faster RCNN	80.7	15.75	177.3	115.684	108.3
YOLOv7	76.0	48.25	51.6	36.490	71.3
YOLO8n	84.5	39.89	8.1	3.006	6.3
YOLO8s	84.0	43.82	28.4	11.127	22.5
YOLO8m	82.5	45.41	78.7	25.841	52.0
YOLO8l	83.6	39.91	164.8	43.609	87.7
YOLO8x	82.4	37.54	257.4	68.126	136.7
YOLOv10n	82.8	43.57	6.0	2.207	5.6
YOLOv10s	81.9	44.49	21.4	7.219	16.5
YOLOv10m	82.5	44.57	58.9	15.315	33.5
YOLOv10l	82.6	37.37	228.7	53.522	107.7
YOLOv10x	79.0	41.60	160.0	29.399	130.4
YOLOv11n	85.9	43.99	6.3	2.583	5.5
YOLOv11s	85.3	43.12	21.3	9.414	19.2
YOLOv11m	85.3	37.89	67.7	20.032	40.5
YOLOv11l	85.9	37.15	194.4	56.830	114.4
YOLOv11x	84.0	37.21	194.4	58.362	114.4
YOLOv12n	84.1	39.71	5.8	2.509	11.2
YOLOv12s	82.0	40.57	19.3	9.074	18.6
YOLOv12m	84.6	41.98	59.5	17.579	39.7
YOLOv12l	79.9	36.23	82.1	26.396	53.7
YOLOv12x	81.3	33.39	184.1	59.248	119.5
YOLOv10 l + SegNeXt	87.5	66.20	114.2	19.765	40.0

表4

表5

YOLOv10不同规模模型在SegNeXt与Lamp模块作用下的检测性能分析

模型名称	mAP 50/%	FPS/（帧/s）	计算量/GFLOPs	模型参数量/MB	权重文件大小/MB
YOLOv10n	82.8	43.57	6.000	2.207	5.6
YOLOv10n+ SegNeXt	85.5	41.50	8.600	2.450	6.3
YOLOv10n+ SegNeXt+Lamp	84.8	51.20	6.500	2.000	5.1
YOLOv10l	82.6	37.37	106.700	45.811	107.7
YOLOv10 l + SegNeXt	87.5	35.80	228.765	53.522	107.6
YOLOv10l+ SegNeXt+Lamp	85.9	66.90	114.252	19.765	40.0

表5

[](https://www.smartag.net.cn/CN/10.12133/j.smartag.SA202507045#f