多源场景下粘虫板小目标害虫轻量化检测识别模型

doi:10.12133/j.smartag.SA202410019

Smart Agriculture ›› 2025, Vol. 7 ›› Issue (1): 111-123.doi: 10.12133/j.smartag.SA202410019

• 专题--农业知识智能服务和智慧无人农场（下） • 上一篇下一篇

多源场景下粘虫板小目标害虫轻量化检测识别模型

杨信廷²^,³, 胡焕¹^,²^,³, 陈晓¹^,²^,³, 李汶政¹^,²^,³, 周子洁²^,³^,⁴, 李文勇²^,³()

^1. 上海海洋大学信息学院，上海 201306，中国
^2. 国家农业信息化工程技术研究中心，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. 吉林农业大学信息技术学院，吉林长春 130118，中国

收稿日期:2024-10-21 出版日期:2025-01-30
基金项目:
国家重点研发计划项目(2022YFD2001804); 北京市农林科学院协同创新中心建设专项
作者简介:
杨信廷，博士，研究员，研究方向为农业信息化关键技术研究。E-mail：yangxt@nercita.org.cn
通信作者:
李文勇，博士，研究员，研究方向为植保信息化。E-mail：liwy@nercita.org.cn

Lightweight Detection and Recognition Model for Small Target Pests on Sticky Traps in Multi-Source Scenarios

YANG Xinting²^,³, HU Huan¹^,²^,³, CHEN Xiao¹^,²^,³, LI Wenzheng¹^,²^,³, ZHOU Zijie²^,³^,⁴, LI Wenyong²^,³()

^1. Shanghai Ocean University, Shanghai 201306, China
^2. National Research Center for Information Technology in Agriculture, Beijing 100097, China
^3. Beijing Academy of Agriculture and Forestry Sciences, Beijing 100097, China
^4. Jilin Agricultural University, Changchun 130118, China

Received:2024-10-21 Online:2025-01-30
Foundation items:National Key Technology R&D Program of China(2022YFD2001804); The Promotion and Innovation of Beijing Academy of Agriculture and Forestry Sciences
About author:
YANG Xinting, E-mail: yangxt@nercita.org.cn
Corresponding author:
LI Wenyong, E-mail: liwy@nercita.org.cn

摘要/Abstract

摘要：

【目的/意义】 为了解决多源场景下粘虫板图像中粉虱和蓟马两种害虫由于个体小难以精确检测以及设备计算资源受限的问题，本研究基于YOLOv5s提出了一种名为MobileNetV4+VN-YOLOv5s的小目标图像轻量化检测识别模型。 【方法】 模型框架结合MobileNetV4主干网络构建EM模块，实现特征提取网络结构的优化和精度的提升；在模型颈部引入轻量化模块GSConv和VoV-GSCSP，替代普通卷积，降低模型复杂度；最后添加NWD（Normalized Wasserstein Distance）损失函数，用于增强小目标的判别敏感度与定位能力。 【结果和讨论】 所提出模型在室内场景下对小目标害虫的检测性能最佳，平均检测精度为82.5%，较原始模型YOLOv5s提升了8.4%；模型参数量降低了3.0 M，帧率提升了6.0帧/s；在室外场景下，所提出模型的平均精度为70.8%，较YOLOv5s提升了7.3个百分点，参数量降低了3.0 M，帧率提升了5.5帧/s；在混合场景下，模型的平均精度为74.7%，较YOLOv5s提升了8.0个百分点，参数量降低了3.0 M，帧率提升了4.3帧/s。同时研究发现，对原始图像进行裁剪分割能够影响模型性能，在采用5×5的比率对原图进行分割下模型的检测识别性能最佳；利用室内场景数据训练的模型能够在所有场景下都获得最佳的检测性能。 【结论】 本研究提出的MobileNetV4+VN-YOLOv5s模型兼顾轻量化和精度，模型可部署到嵌入式设备，便于落地应用，可为各种多源场景下粘虫板图像中小目标害虫检测提供参考。

关键词: 小目标, 害虫检测, 轻量化, 粘虫板, 多源场景, MobileNetV4, YOLOv5s

Abstract:

[Objective] In crop cultivation and production, pests have gradually become one of the main issues affecting agricultural yield. Traditional models often focus on achieving high accuracy, however, to facilitate model application, lightweighting is necessary. The targets in yellow sticky trap images are often very small with low pixel resolution, so modifications in network structure, loss functions, and lightweight convolutions need to adapt to the detection of small-object pests. Ensuring a balance between model lightweighting and small-object pest detection is particularly important. To improve the detection accuracy of small target pests on sticky trap images from multi-source scenarios, a lightweight detection model named MobileNetV4+VN-YOLOv5s was proposed in this research to detect two main small target pests in agricultural production, whiteflies and thrips. [Methods] In the backbone layer of MobileNetV4+VN-YOLOv5s, an EM block constructed with the MobileNetV4 backbone network was introduced for detecting small, high-density, and overlapping targets, making it suitable for deployment on mobile devices. Additionally, the Neck layer of MobileNetV4+VN-YOLOv5s incorporates the GSConv and VoV-GSCSP modules to replace regular convolutional modules with lightweight design, effectively reducing the parameter size of the model while improving detection accuracy. Lastly, a normalized wasserstein distance (NWD)loss function was introduced into the framework to enhance the sensitivity for low-resolution small target pests. Extensive experiments including state-of-the-art comparison, ablation evaluation, performance analysis on image splitting, pest density and multi-source data were conducted. [Results and Discussions] Through ablation tests, it was concluded that the EM module and the VoV-GSCSP convolution module had significant effects in reducing the model parameter size and frame rate, the NWD loss function significantly improved the mean average precision (mAP) of the model. By comparing tests with different loss functions, the NWD loss function improves the mAP by 6.1, 10.8 and 8.2 percentage compared to the DIoU, GIoU and EIoU loss functions, respectively, so the addition of the NWD loss function achieved good results. Comparative performance tests were detected wiht different light weighting models, the experimental results showed that the mAP of the proposed MobileNetV4+VN-YOLOv5s model in three scenarios (Indoor, Outdoor, Indoor&Outdoor) was 82.5%, 70.8%, and 74.7%, respectively. Particularly, the MobileNetV4+VN-YOLOv5s model had a parameter size of only 4.2 M, 58% of the YOLOv5s model, the frame rate was 153.2 fps, an increase of 6.0 fps compared to the YOLOv5s model. Moreover, the precision and mean average precision reach 79.7% and 82.5%, which were 5.6 and 8.4 percentage points higher than the YOLOv5s model, respectively. Comparative tests were conducted in the upper scenarios based on four splitting ratios: 1×1, 2×2, 5×5, and 10×10. The most superior was the result by using 5×5 ratio in indoor scenario, and the mAP of this case reached 82.5%. The mAP of the indoor scenario was the highest in the low-density case, reaching 83.8%, and the model trained based on the dataset from indoor condition achieves the best performance. Comparative tests under different densities of pest data resulted in a decreasing trend in mAP from low to high densities for the MobileNetV4+VN-YOLOv5s model in the three scenarios. Based on the comparison of the experimental results of different test sets in different scenarios, all three models achieved the best detection accuracy on the IN dataset. Specifically, the IN-model had the highest mAP at 82.5%, followed by the IO-model. At the same time, the detection performance showed the same trend across all three test datasets: The IN model performed the best, followed by the IO-model, and the OUT-model performed the lowest. By comparing the tests with different YOLO improvement models, it was concluded that MobileNetV4+VN-YOLOv5s had the highest mAP, EVN-YOLOv8s was the second highest, and EVN-YOLOv11s was the lowest. Besides, after deploying the model to the Raspberry Pi 4B motherboard, it was concluded that the detection results of the YOLOv5s model had more misdetections and omissions than those of the MobileNetV4+VN-YOLOv5s model, and the time of the model was shortened by about 33% compared to that of the YOLOv5s model, which demonstrated that the model had a good prospect of being deployed in the application. [Conclusions] The MobileNetV4+VN-YOLOv5s model proposed in this study achieved a balance between lightweight design and accuracy. It can be deployed on embedded devices, facilitating practical applications. The model can provide a reference for detecting small target pests in sticky trap images under various multi-source scenarios.

Key words: small target, pest detection, lightweight, sticky trap, multi-source scenarios, MobileNetV4, YOLOv5s

中图分类号:

S433
TP391.4

杨信廷, 胡焕, 陈晓, 李汶政, 周子洁, 李文勇. 多源场景下粘虫板小目标害虫轻量化检测识别模型[J]. 智慧农业(中英文), 2025, 7(1): 111-123.

YANG Xinting, HU Huan, CHEN Xiao, LI Wenzheng, ZHOU Zijie, LI Wenyong. Lightweight Detection and Recognition Model for Small Target Pests on Sticky Traps in Multi-Source Scenarios[J]. Smart Agriculture, 2025, 7(1): 111-123.

图/表 22

图1

表1

图2

图3

图4

图5

图6

图7

图8

图9

表2

表3

表4

表5

图10

表6

图11

表7

图12

表8

图13

图14

参考文献 31

1	吕秀英. 农业小虫蓟马的危害与综合防控技术[J]. 现代农村科技, 2023(9): 31-32.
	LYU X Y. Hazards of agricultural insect thrips and integrated prevention and control technology[J]. Modern Rural Science and Technology, 2023, (9): 31-2.
2	罗宏伟, 冯钦, 王建波, 等. 小巢粉虱的危害程度调查[J]. 农业科技通讯, 2022(1): 137-139.
	LUO H W, FENG Q, WANG J B, et al. Investigation on the damage degree of whitefly in nest[J]. Bulletin of agricultural science and technology, 2022(1): 137-139.
3	QI J T, LIU X N, LIU K, et al. An improved YOLOv5 model based on visual attention mechanism: Application to recognition of tomato virus disease[J]. Computers and electronics in agriculture, 2022, 194: ID 106780.
4	REN S Q, HE K M, 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, 39(6): 1137-1149.
5	LI W Y, WANG D J, LI M, et al. Field detection of tiny pests from sticky trap images using deep learning in agricultural greenhouse[J]. Computers and electronics in agriculture, 2021, 183: ID 106048.
6	LI W Y, YANG Z K, LV J W, et al. Detection of small-sized insects in sticky trapping images using spectral residual model and machine learning[J]. Frontiers in plant science, 2022, 13: ID 915543.
7	ZHANG H B, QIN L F, LI J, et al. Real-time detection method for small traffic signs based on YOLOv3[J]. IEEE access, 2020, 8: 64145-64156.
8	WANG D J, WANG Y Z, LI M, et al. Using an improved YOLOv4 deep learning network for accurate detection of whitefly and Thrips on sticky trap images[J]. Transactions of the ASABE, 2021, 64(3): 919-927.
9	CHENG Z K, HUANG R Q, QIAN R, et al. A lightweight crop pest detection method based on convolutional neural networks[J]. Applied sciences, 2022, 12(15): ID 7378.
10	ZHANG S, WANG H, ZHANG C, et al. JutePest-YOLO: A deep learning network for jute pest identification and detection[J]. IEEE access, 2024, 12: 72938-72956.
11	XIANG Q C, HUANG X N, HUANG Z X, et al. YOLO-Pest: An insect pest object detection algorithm via CAC3 module[J]. Sensors, 2023, 23(6): ID 3221.
12	LI D W, AHMED F, WU N L, et al. YOLO-JD: A deep learning network for jute diseases and pests detection from images[J]. Plants, 2022, 11(7): ID 937.
13	DONG Q, SUN L N, HAN T X, et al. PestLite: A novel YOLO-based deep learning technique for crop pest detection[J]. Agriculture, 2024, 14(2): ID 228.
14	XUE Z Y, XU R J, BAI D, et al. YOLO-tea: A tea disease detection model improved by YOLOv5[J]. Forests, 2023, 14(2): ID 415.
15	QIN D F, LEICHNER C, DELAKIS M, et al. MobileNetV4: universal models for the mobile ecosystem[C]// Computer Vision – ECCV 2024. Cham, Germany: Springer Nature Switzerland, 2024: 78-96.
16	LI H, LI J, WEI H, et al. Slim-neck by GSConv: A better design paradigm of detector architectures for autonomous vehicles[EB/OL]. arXiv: 220602424, 2022.
17	WANG J W, XU C, YANG W, et al. A normalized gaussian wasserstein distance for tiny object detection[EB/OL]. arXiv: 211013389, 2023.
18	张博, 张苗辉, 陈运忠. 基于空间金字塔池化和深度卷积神经网络的作物害虫识别[J]. 农业工程学报, 2019, 35(19): 209-215.
	ZHANG B, ZHANG M H, CHEN Y Z. Crop pest identification based on spatial pyramid pooling and deep convolution neural network[J]. Transactions of the Chinese society of agricultural engineering, 2019, 35(19): 209-215.
19	朱德利, 文瑞, 熊俊逸. 融合坐标注意力机制的轻量级玉米花丝检测[J]. 农业工程学报, 2023, 39(3): 145-153.
	ZHU D L, WEN R, XIONG J Y. Lightweight corn silk detection network incorporating with coordinate attention mechanism[J]. Transactions of the Chinese society of agricultural engineering, 2023, 39(3): 145-153.
20	FU Z L, YIN L F, CUI C, et al. A lightweight MHDI-DETR model for detecting grape leaf diseases[J]. Frontiers in plant science, 2024, 15: ID 1499911.
21	王泽钧, 马凤英, 张瑜, 等. 基于注意力机制和多尺度轻量型网络的农作物病害识别[J]. 农业工程学报, 2022, 38(Z): 176-183.
	WANG Z J, MA F Y, ZHANG Y, et al. Crop disease recognition using attention mechanism and multi-scale lightweight network [J]. Transactions of the Chinese society of agricultural engineering, 2022,38(Z): 176-83.
22	TIAN Y J, SU D, LAURIA S, et al. Recent advances on loss functions in deep learning for computer vision[J]. Neurocomputing, 2022, 497(C): 129-158.
23	PANARETOS V M, ZEMEL Y. Statistical aspects of Wasserstein distances[J]. Annual review of statistics and its application, 2019, 6: 405-431.
24	REZATOFIGHI H, TSOI N, GWAK J, et al. Generalized intersection over union: A metric and a loss for bounding box regression[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, New Jersey, USA: IEEE, 2019: 658-666.
25	ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: Faster and better learning for bounding box regression[J]. Proceedings of the AAAI conference on artificial intelligence, 2020, 34(7): 12993-13000.
26	YANG Z M, WANG X L, LI J G. EIoU: An improved vehicle detection algorithm based on VehicleNet neural network[J]. Journal of physics: Conference series, 2021, 1924(1): ID 012001.
27	MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2: Practical guidelines for efficient CNN architecture design[C]// Computer Vision-ECCV 2018. Cham, Germany: Springer International Publishing, 2018: 122-138.
28	HAN K, WANG Y H, TIAN Q, et al. GhostNet: More features from cheap operations[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, New Jersey, USA: IEEE, 2020: 1580-1589.
29	KOONCE B. Convolutional neural networks with swift for tensorflow: Image recognition and dataset categorization[M]. Berkeley, CA: Apress, 2021
30	REIS D, KUPEC J, HONG J, et al. Real-time flying object detection with YOLOv8[EB/OL]. arXiv: 230509972, 2023.
31	RAHIMA KHANAM M H. YOLOv11: An overview of the key architectural enhancements[EB/OL]. arXiv: 241017725, 2024.

场景	数据集	类型	图像数量/张	小图数量/张	小图大小/pixels	粉虱数量/只	蓟马数量/只
室内	IN-Dataset	训练集	48	1 200	800×600	5 567	3 743
		验证集	6	150	800×600	1 743	1 247
		测试集	6	150	800×600	623	2 096
室外	OUT-Dataset	训练集	48	1 200	800×600	10 713	2 845
		验证集	6	150	800×600	1 659	1 212
		测试集	6	150	800×600	1 545	1 176
混合	IO-Dataset	训练集	48	1 200	800×600	11 832	3 471
		验证集	6	150	800×600	1 936	1 517
		测试集	6	150	800×600	1 079	1 170

模型	EM block	VoV- GSCSP	Loss function	精确度/%	召回率/%	mAP@0.5/%	参数量/M	帧率/（帧/s）
YOLOv5s	—	—	—	74.1	71.3	74.1	7.2	147.2
E-YOLOv5s	√	—	—	77.3	76.9	78.2	6.8	148.1
V-YOLOv5s	—	√	—	73.8	72.7	74.5	5.4	152.9
N-YOLOv5s	—	—	√	77.2	77.5	78.7	7.2	142.7
EV-YOLOv5s	√	√	—	75.6	76.2	77.4	4.2	156.6
EN-YOLOv5s	√	—	√	79.4	79.7	81.2	6.8	146.5
VN-YOLOv5s	—	√	√	76.9	77.4	78.5	5.4	150.3
Our model	√	√	√	79.7	80.4	82.5	4.2	153.2

损失函数	精确度/%	召回率/%	mAP@0.5/%	帧率/（帧/s）
DIoU	74.9	75.5	76.4	151.7
GIoU	70.6	70.2	71.7	152.3
EIoU	72.8	73.5	74.3	148.5
NWD	79.7	80.4	82.5	153.2

模型	IN-Dataset			OUT-Dataset			IO-Dataset
模型	mAP@0.5/%	参数量/M	帧率/（帧/s）	mAP@0.5/%	参数量/M	帧率/（帧/s）	mAP@0.5/%	参数量/M	帧率/（帧/s）
YOLOv5s	74.1	7.2	147.2	63.5	7.2	134.8	66.7	7.2	139.9
ShuffleNetV2+YOLOv5s	72.6	4.7	153.7	64.6	4.7	139.6	64.9	4.7	146.4
GhostNet+YOLOv5s	74.8	5.5	144.6	62.2	5.5	131.5	65.8	5.5	142.3
MobileNetV3+YOLOv5s	75.6	5.3	143.8	68.1	5.3	133.1	69.4	5.3	136.5
Our model	82.5	4.2	153.2	70.8	4.2	140.3	74.7	4.2	144.2

分割比	图像数量/张	小图数量/张	小图大小/ pixels	室内（mAP@0.5/%）	室外（mAP@0.5/%）	混合（mAP@0.5/%）
1×1	60	60	4 000×3 000	68.9	56.7	64.4
2×2	60	240	2 000×1 500	65.6	54.1	63.5
5×5	60	1 500	800×600	82.5	70.8	74.7
10×10	60	6 000	400×300	73.2	64.9	71.3

多源场景下粘虫板小目标害虫轻量化检测识别模型

Lightweight Detection and Recognition Model for Small Target Pests on Sticky Traps in Multi-Source Scenarios

在线阅读

知网下载

本地下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 22

参考文献 31

相关文章 12

编辑推荐

Metrics

本文评价

场景	密度	测试图数量/张	精确度/%	召回率/%	mAP@0.5/%
室内	低密度	50	81.9	82.2	83.8
	中密度	50	80.2	79.8	81.5
	高密度	50	73.7	74.2	75.4
室外	低密度	50	74.2	73.3	74.5
	中密度	50	71.6	70.5	72.4
	高密度	50	69.1	70.4	70.9
混合	低密度	50	76.5	75.8	77.1
	中密度	50	75.3	74.9	76.2
	高密度	50	70.2	70.6	71.8

模型	训练集	测试集（mAP@0.5/%）
模型	训练集	IN-Dataset-Test set	OUT-Dataset-Test set	IO-Dataset-Test set
IN-Model	IN-Dataset-Train set	82.5	77.2	80.5
OUT-Model	OUT-Dataset-Train set	75.7	70.8	74.3
IO-Model	IO-Dataset-Train set	78.4	75.1	74.7

[1]	吴六爱, 许雪珂. 基于改进YOLOv10n的轻量化番茄叶片病虫害检测方法[J]. 智慧农业(中英文), 2025, 7(1): 146-155.
[2]	靳学萌, 梁西银, 邓鹏飞. 基于改进YOLOv10的轻量级黄花菜分级检测模型[J]. 智慧农业(中英文), 2024, 6(5): 108-118.
[3]	刘伊, 张彦军. ReluformerN：轻量化高低频增强高光谱农业地物分类方法[J]. 智慧农业(中英文), 2024, 6(5): 74-87.
[4]	叶大鹏, 景均, 张之得, 李辉煌, 吴昊宇, 谢立敏. MSH-YOLOv8：融合尺度重建的蘑菇小目标检测方法[J]. 智慧农业(中英文), 2024, 6(5): 139-152.
[5]	胡程喜, 谭立新, 王文胤, 宋敏. 基于改进DeepLabV3+的轻量化茶叶嫩芽采摘点识别模型[J]. 智慧农业(中英文), 2024, 6(5): 119-127.
[6]	翁智, 刘海鑫, 郑志强. CSD-YOLOv8s：基于无人机图像的密集小目标羊只检测模型[J]. 智慧农业(中英文), 2024, 6(4): 42-52.
[7]	张荣华, 白雪, 樊江川. 复杂场景下害虫目标检测算法：YOLOv8-Extend[J]. 智慧农业(中英文), 2024, 6(2): 49-61.
[8]	杨锋, 姚晓通. 基于改进YOLOv8的小麦叶片病虫害检测轻量化模型[J]. 智慧农业(中英文), 2024, 6(1): 147-157.
[9]	夏雪, 柴秀娟, 张凝, 周硕, 孙琦鑫, 孙坦. 用于边缘计算设备的果树挂果量轻量化估测模型[J]. 智慧农业(中英文), 2023, 5(2): 1-12.
[10]	罗庆, 饶元, 金秀, 江朝晖, 王坦, 王丰仪, 张武. 基于改进YOLOv5s和多模态图像的树上毛桃检测[J]. 智慧农业(中英文), 2022, 4(4): 84-104.
[11]	夏烨, 雷哓晖, 祁雁楠, 徐陶, 袁全春, 潘健, 姜赛珂, 吕晓兰. 基于改进Ghost-YOLOv5s-BiFPN算法检测梨树花序[J]. 智慧农业(中英文), 2022, 4(3): 108-119.
[12]	郭秀明, 诸叶平, 李世娟, 张杰, 吕纯阳, 刘升平. 农业复杂环境下尺度自适应小目标识别算法——以蜜蜂为研究对象[J]. 智慧农业(中英文), 2022, 4(1): 140-149.