基于YOLOv5s的轻量级茶叶嫩芽终端检测模型

doi:10.3969/j.issn.1004-1524.20230822

浙江农业学报 ›› 2024, Vol. 36 ›› Issue (6): 1413-1424.DOI: 10.3969/j.issn.1004-1524.20230822

基于YOLOv5s的轻量级茶叶嫩芽终端检测模型

朱铭敏¹(), 张国平¹^,^*(), 谭建军², 孙玲姣², 朱黎¹^,³, 焦洁²

1.华中师范大学物理科学与技术学院,湖北武汉 430079
2.湖北恩施学院智能工程学院,湖北恩施 445000
3.湖北民族大学智能科学与工程学院,湖北恩施 445000

收稿日期:2023-07-03 出版日期:2024-06-25 发布日期:2024-07-02
作者简介:朱铭敏(1999—),女,湖北宜昌人,硕士研究生,主要从事人工智能研究。E-mail:892359179@qq.com
通讯作者: *张国平,E-mail: gpzhang@ccnu.edu.cn
基金资助:
湖北恩施学院研究生联合培养项目(KYYL202304);国家自然科学基金地区科学基金项目(61961017);湖北省中央引导地方科技发展专项(ZYYD2022000156);湖北省恩施州科技计划(D20220004)

A lightweight tea buds terminal detection model based on YOLOv5s

ZHU Mingmin¹(), ZHANG Guoping¹^,^*(), TAN Jianjun², SUN Lingjiao², ZHU Li¹^,³, JIAO Jie²

1. College of Physical Science and Technology, Central China Normal University, Wuhan 430079, China
2. College of Intelligent Engineering, Hubei Enshi College, Enshi 445000, Hubei,China
3. College of Intelligent Science and Engineering, Hubei Minzu University, Enshi 445000, Hubei, China

Received:2023-07-03 Online:2024-06-25 Published:2024-07-02

摘要/Abstract

摘要：

在茶园环境中快速精准识别茶叶嫩芽是实现智能化采茶的关键技术之一,但茶芽检测模型的复杂性导致模型参数量大、计算量大、模型尺寸大,限制了模型在采茶机器人嵌入式设备的部署。鉴于此,本文提出一种基于YOLOv5s的轻量级茶叶嫩芽终端检测模型。首先,使用轻量级网络GhostNet替换YOLOv5s中的Backbone网络,并重构Neck网络,降低模型的参数量、计算量和内存占用量,改进后的模型分别降低了47.64%、49.36%、45.51%。其次,通过引入协调注意力(coordinate attention, CA)机制,抑制图像背景信息,增强模型对茶叶嫩芽的特征提取能力。接着,在Neck网络引入多尺度特征融合(multi-scale context, MSC)模块,有效融合浅层图像特征和深层语义特征,帮助网络模型提取有效识别信息。最后,使用边界框回归损失函数EIOU替换CIOU,加快损失函数收敛速度,提高茶叶嫩芽边界框定位精度。试验结果表明,与原YOLOv5s模型相比,改进模型的参数量、计算量以及模型内存占用量分别降低了3 Mb、7.3 Gb和6.37 Mb,检测精度提升0.3%。通过模型转换将该模型移植到树莓派平台,经过环境部署和推理引擎加速,达到了轻量级模型在资源和算力有限的树莓派上对茶叶嫩芽检测的目的,在一定程度上提高了茶叶嫩芽的识别精确度,为茶叶嫩芽的智能化采摘提供了理论研究和技术支持。

关键词: 茶叶嫩芽检测, 树莓派, 轻量级, 注意力机制

Abstract:

Rapid and accurate identification of tea buds in tea garden environments is one of the key technologies for achieving intelligent tea picking. However, the complexity of the tea buds detection model leads to problems such as large model parameters, computational complexity, and model size, which limits the deployment of this model in embedded devices of tea picking robots. In view of this, this article proposes a lightweight tea buds terminal detection model based on YOLOv5s. Firstly, the lightweight network GhostNet is used to replace the Backbone network in YOLOv5s, and the Neck network is reconstructed to reduce the parameters, computation and memory consumption of the model. The improved model reduces 47.64%, 49.36% and 45.51% respectively. Secondly, by introducing a coordinated attention(CA) mechanism to suppress image background information, the model’s feature extraction ability for tea buds is enhanced. Next, multi-scale context (MSC) module is introduced into the Neck network to effectively fuse shallow image features and deep semantic features, which helps the network model extract effective recognition information. Then, the boundary box regression Loss function CIOU is replaced by EIOU to accelerate the Rate of convergence of the Loss function and improve the positioning accuracy of the tea buds boundary box. The experiment result shows that compared with the original YOLOv5s model, the improved model reduces the parameter count, computational complexity, and model memory usage by 3 Mb, 7.3 Gb, and 6.37 Mb, respectively, and improves detection accuracy by 0.3%. Finally, the model was transplanted to the Raspberry Pi platform through model transformation. After environmental deployment and inference engine acceleration, the lightweight model achieved the goal of detecting tea buds on Raspberry Pi with limited resources and computing power. It also improved the recognition accuracy of tea buds to a certain extent, providing theoretical research and technical support for the intelligent picking of tea buds.

Key words: tea bud detection, Raspberry Pie, lightweight, attention mechanism

中图分类号:

S24
TP391.4

朱铭敏, 张国平, 谭建军, 孙玲姣, 朱黎, 焦洁. 基于YOLOv5s的轻量级茶叶嫩芽终端检测模型[J]. 浙江农业学报, 2024, 36(6): 1413-1424.

ZHU Mingmin, ZHANG Guoping, TAN Jianjun, SUN Lingjiao, ZHU Li, JIAO Jie. A lightweight tea buds terminal detection model based on YOLOv5s[J]. Acta Agriculturae Zhejiangensis, 2024, 36(6): 1413-1424.

图/表 16

图1 原始图像及数据增强示例

Fig.1 The original images and images after data enhancement

图2 YOLOv5s结构图 Conv2d,二维卷积;BN,批归一化;SiLU,激活函数;CBS,卷积块;Bottleneck,瓶颈结构;C3,轻量化语义分割网络;SPPF,快速空间金字塔池化模块;Concat,张量拼接;Upsample,上采样,MaxPool,最大池化。

Fig.2 Structure diagram of YOLOv5s Conv2d, Two-dimensional convolution; BN, Batch normalization; SiLU, Activation function; CBS, Convolutional block; Bottleneck, Bottleneck structure; C3, Lightweight semantic segmentation network; SPPF, Fast spatial pyramid pooling module; Concat, Tensor splicing; Upsample, Upsampling; MaxPool, Maximum pooling.

图3 轻量级卷积模块(ghost module)的结构

Fig.3 Structure diagram of ghost module

图4 幻象瓶颈层的结构 DWConv,深度可分离卷积;Stride,步长。

Fig.4 Structure diagram of ghost bottleneck DWConv, Depth separable convolution; Stride, Step size.

图5 协调注意力(CA)机制的结构 Residual,残差模块;AvgPooling,平均池化;BatchNorm,批归一化;Sigmoid,激活函数;Re-weight,重加权模块。

Fig.5 Structure diagram of coordinate attention (CA) mechanism Residual, Residual module; AvgPooling, Average pooling; BatchNorm, Batch normalization; Sigmaid, Activation function; Re weight, Reweighting module.

图6 多尺度特征融合(MSC)的结构示意图

Fig.6 Structure diagram of multi-scale context

图7 改进后的YOLOv5s网络的结构 Ghost CBS,Ghost卷积块;Ghost C3, Ghost轻量化语义分割网络;DWConv,深度可分离卷积;CA,注意力机制;MSC,多尺度特征融合模块;Ghost Bottleneck, 幻象瓶颈层。

Fig.7 Structure diagram of the improved YOLOv5s network Ghost CBS, Ghost convolutional block; Ghost C3, Ghost lightweight semantic segmentation network; DWConv, Depth separable convolution; CA, Attention mechanism; MSC, Multi-scale feature fusion module; Ghost Bottleneck, Ghost Bottleneck.

图8 树莓派4B的实物图

Fig.8 Photos of Raspberry Pie 4B

图9 ONNXRuntime执行流程

Fig.9 The execution process of ONNXRuntime

表1 不同轻量化模型对比

Table 1 Comparison of different lightweight models

模型 Models	参数量 Params/Mb	计算量 FLOPs/Gb	模型大小 Size/Mb	精准率 P/%	召回率 R/%	平均精度 AP/%
MobileNetV3	3.54	7.0	7.5	62.2	64.0	65.2
ShuffleNetV2	3.68	7.8	7.6	66.3	66.5	68.4
GhostNet	3.67	8.0	7.9	67.0	68.2	71.2

表2 消融试验对比

Table 2 Comparison of ablation test results

模型 Models	参数量 Params/Mb	计算量 FLOPs/Gb	模型大小 Size/Mb	精准率 P/%	召回率 R/%	平均精度 AP/%
YOLOv5s	7.01	15.8	14.50	68.8	67.5	72.7
A	3.67	8.0	7.90	67.0	68.2	71.2
B	3.74	8.1	8.00	67.5	68.4	72.0
C	3.98	8.5	8.13	68.4	68.5	72.5
D	4.01	8.5	8.13	68.5	68.8	73.0

表3 不同检测算法模型对比

Table 3 Comparison of different detection algorithm models

模型 Models	参数量 params/Mb	计算量 FLOPs/Gb	模型大小 size/Mb	精准率 P/%	召回率 R/%	平均精度 AP/%	t/s
Faster R-CNN^[28]	—	—	—	89.00	58.00	54.00	—
YOLOv3^[29]	—	—	—	74.51	69.56	71.96	—
Compact-YOLO v4^[30]	—	—	23.20	51.07	78.67	72.93	0.023
YOLOv5-Lite	1.54	3.70	3.40	60.30	61.70	60.00	0.063
YOLOv5s	7.01	15.8	14.50	68.80	67.50	72.70	0.014
Ours	4.01	8.50	8.13	68.50	68.80	73.00	0.016

图10 YOLOv5s模型改进前(A)、后(B)的结果对比

Fig.10 Comparison of results before (A) and after (B) improvement of YOLOv5s model

图11 树莓派上模型改进前(A)、后(B)的结果对比

Fig.11 Comparison of results before (A) and after (B) model improvement on Raspberry Pie

图12 ONNXRuntime推理前(A)、后(B)结果对比

Fig.12 Comparison of results before (A) and after (B) ONNXRuntime inference

表4 模型检测耗时

Table 4 Detection time of different models s

模型 Models	t₁	t₂	t₃	t₄
YOLOv5s	0.014	0.155	16.572	0.892
Ours	0.016	0.145	9.157	0.723

参考文献 30

[1]	刘顺星, 张慧坚, 刘晓青. 中国茶产业国际竞争力的比较分析与提升对策[J/OL]. 中国食物与营养, 2023: 1-8.
	LIU S X, ZHANG H J, LIU X Q. Comparative analysis and promotion countermeasures of international competitiveness of Chinese tea industry[J/OL]. Food and Nutrition in China, 2023: 1-8. (in Chinese with English abstract)
[2]	SHAO P D, WU M H, WANG X W, et al. Research on the tea bud recognition based on improved k-means algorithm[J]. MATEC Web of Conferences, 2018, 232: 03050.
[3]	刘自强, 周铁军, 傅冬和, 等. 基于颜色和形状的鲜茶叶图像特征提取及在茶树品种识别中的应用[J]. 江苏农业科学, 2021, 49(12): 168-172.
	LIU Z Q, ZHOU T J, FU D H, et al. Color and shape based feature extraction of fresh tea image and its application in tea tree variety recognition[J], Jiangsu Agricultural Sciences, 2021, 49(12): 168-172. (in Chinese with English abstract)
[4]	吴雪梅, 张富贵, 吕敬堂. 基于图像颜色信息的茶叶嫩叶识别方法研究[J]. 茶叶科学, 2013, 33(6): 584-589.
	WU X M, ZHANG F G, LV J T. Research on recognition of tea tender leaf based on image color information[J]. Journal of Tea Science, 2013, 33(6): 584-589. (in Chinese with English abstract)
[5]	OTSU N. A threshold selection method from gray-level histograms[J]. IEEE Transactions on Systems, Man, and Cybernetics, 1979, 9(1): 62-66.
[6]	王子钰, 赵怡巍, 刘振宇. 基于SSD算法的茶叶嫩芽检测研究[J]. 微处理机, 2020, 41(4): 42-48.
	WANG Z Y, ZHAO Y W, LIU Z Y. Research on tea buds detection based on SSD algorithmon[J]. Microprocessors, 2020, 41(4): 42-48. (in Chinese with English abstract)
[7]	GAO M, SHI M, LI C. Research and implementation of image recognition of tea based on deep learning[C]// ACIS International Winter Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD-Winter), January 28-30, 2021. Ho Chi Minh City, Vietnam. IEEE, 2021: 63-68.
[8]	FUKUSHIMA K. Neural network model for a mechanism of patten recognition unaffected by shift in position - Neocognitron[J]. IEICE Technical Report, 1989, 62(10): 658-665. (in Chinese with English abstract)
[9]	FUKUSHIMA K. Neocognitron: a self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position[J]. Biological Cybernetics, 1980, 36(4): 193-202.
[10]	XU W K, ZHAO L G, LI J, et al. Detection and classification of tea buds based on deep learning[J]. Computers and Electronics in Agriculture, 2022, 192: 106547.
[11]	GONG T, WANG Z L. A tea tip detection method suitable for tea pickers based on YOLOv4 network[C]// International Symposium on Robotics & Intelligent Manufacturing Technology (ISRIMT). September 24-26, 2021. Changzhou, China. IEEE, 2021: 264-268.
[12]	尹川, 苏议辉, 潘勉, 等. 基于改进YOLOv5s的名优绿茶品质检测[J]. 农业工程学报, 2023, 39(8): 179-187.
	YIN C, SU Y H, PAN M, et al. Detection of the quality of famous green tea based on improved YOLOv5s[J]. Transactions of the Chinese Society of Agricultural Engineering, 2023, 39(8): 179-187. (in Chinese with English abstract)
[13]	GUI Z Y, CHEN J N, LI Y, et al. A lightweight tea bud detection model based on Yolov5[J]. Computers and Electronics in Agriculture, 2023, 205: 107636.
[14]	刘翀豪, 潘理虎, 杨帆, 等. 改进YOLOv5的轻量化口罩检测算法[J]. 计算机工程与应用, 2023, 59(7): 232-241.
	LIU C H, PAN L H, YANG F, et al. Improved YOLOv5 lightweight mask detection algorithm[J]. Computer Engineering and Applications, 2023, 59(7): 232-241. (in Chinese with English abstract)
[15]	Howard A, Ruoming Pang, Adam H, et al. Searching for MobileNetV3[C]// IEEE/CVF International Conference on Computer Vision (ICCV). October 27-November 2, 2019. Seoul, Korea (South). IEEE, 2019: 1314-1324.
[16]	HAN K, WANG Y H, TIAN Q, et al. GhostNet: More Features From Cheap Operations[C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 13-19, 2020. Seattle, WA, USA. IEEE, 2020: 1577-1586.
[17]	WANG L, LI B, FU R G. Feature fusion and coordinate attention for small target detection[C]// International Conference on Intelligent Computing and Signal Processing (ICSP). April 15-17, 2022. Xi’an, China. IEEE, 2022: 629-633.
[18]	ZHANG Z, DENG Z J, WU Z P, et al. An improved EIoU-Yolov5 algorithm for blood cell detection and counting[C]// International Conference on Pattern Recognition and Artificial Intelligence (PRAI). August 19-21, 2022. Chengdu, China. IEEE, 2022: 989-993.
[19]	XUE J L, CHENG F, LI Y Q, et al. Detection of farmland obstacles based on an improved YOLOv5s algorithm by using CIoU and anchor box scale clustering[J]. Sensors, 2022, 22(5): 1790.
[20]	ZHANG Y M, HSIEH J W, LEE C C, et al. SFPN: synthetic FPN for object detection[C]// IEEE International Conference on Image Processing (ICIP). October 16-19, 2022. Bordeaux, France. IEEE, 2022: 1316-1320.
[21]	LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition. June 18-23, 2018. Salt Lake City, UT. IEEE, 2018: 8759-8768.
[22]	CAO M L, FU H, ZHU J Y, et al. Lightweight tea bud recognition network integrating GhostNet and YOLOv5[J]. Mathematical Biosciences and Engineering, 2022, 19(12): 12897-12914.
[23]	LU X, TAO M Y, FU X, et al. Lightweight network design based on ResNet structure for modulation recognition[C]// 2021 IEEE 94th Vehicular Technology Conference (VTC2021-Fall). September 27-30, 2021. Norman, OK, USA. IEEE, 2021: 1-5.
[24]	李晨, 张辉, 张邹铨, 等. 融合多尺度特征与全局上下文信息的X光违禁物品检测[J]. 中国图象图形学报, 2022, 27(10): 3043-3057.
	LI C, ZHANG H, ZHANG Z Q, et al. Integrated multi-scale features and global context in X-ray detection for prohibited items[J]. Journal of Image and Graphics, 2022, 27(10): 3043-3057. (in Chinese with English abstract)
[25]	GUO S Y, LI Z Y, GUO S W. Pulmonary nodule detection algorithm based on YOLOv3[C]// 2022 4th International Conference on Artificial Intelligence and Advanced Manufacturing (AIAM). October 7-9, 2022. Hamburg, Germany. IEEE, 2022: 677-679.
[26]	HUA Y H, XIAO B G, LI J Y, et al. Design and implementation of face recognition system based on raspberry pie[C]// 2021 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech). October 25-28, 2021. AB, Canada. IEEE, 2021: 673-678.
[27]	RAN H H, WEN S P, WANG S Q, et al. Memristor-based edge computing of ShuffleNetV2 for image classification[J]. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2021, 40(8): 1701-1710.
[28]	朱红春, 李旭, 孟炀, 等. 基于Faster R-CNN网络的茶叶嫩芽检测[J]. 农业机械学报, 2022, 53(5): 217-224.
	ZHU H C, LI X, MENG Y, et al. Tea bud detection based on faster R-CNN network[J]. Transactions of the Chinese Society for Agricultural Machinery, 2022, 53(5): 217-224. (in Chinese with English abstract)
[29]	CHEN C L, LU J Z, ZHOU M C, et al. A YOLOv3-based computer vision system for identification of tea buds and the picking point[J]. Computers and Electronics in Agriculture, 2022, 198(1): 107116.
[30]	黄家才, 唐安, 陈光明, 等. 基于Compact-YOLO v4的茶叶嫩芽移动端识别方法[J]. 农业机械学报, 2023, 54(3): 282-290.
	HUANG J C, TANG A, CHEN G M, et al. Mobile recognition solution of tea buds based on compact-YOLO v4 algorithm[J]. Transactions of the Chinese Society for Agricultural Machinery, 2023, 54(3): 282-290. (in Chinese with English abstract)

编辑推荐

Metrics

阅读次数

全文

376

HTML			PDF

最新录用	在线预览	正式出版	最新录用	在线预览	正式出版
0	0	8	0	0	368

来源	本网站	其他网站

次数	279	97
比例	74%	26%

摘要

147

最新录用	在线预览	正式出版

0	0	147

	来源	本网站

	次数	147
	比例	100%

基于YOLOv5s的轻量级茶叶嫩芽终端检测模型

A lightweight tea buds terminal detection model based on YOLOv5s

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 16

参考文献 30

相关文章 6

编辑推荐

Metrics

本文评价

[1]	李大华, 孔舒, 李栋, 于晓. 基于改进SSD模型的柑橘叶片病害轻量化检测模型[J]. 浙江农业学报, 2024, 36(3): 662-670.
[2]	李颀, 李煜哲. 弱光条件下猪舍清洗目标检测[J]. 浙江农业学报, 2023, 35(9): 2240-2249.
[3]	李荣鹏, 买买提·沙吾提, 盛艳芳, 何旭刚. 基于CA-MobileNet-V2的核桃病害识别与应用[J]. 浙江农业学报, 2023, 35(12): 2977-2987.
[4]	王英允, 龙燕, 杨智优, 黄铝文. 基于改进U-Net网络的苹果叶部病害语义分割方法[J]. 浙江农业学报, 2023, 35(11): 2731-2741.
[5]	周品志, 裴悦琨, 魏冉, 张永飞, 谷宇. 基于YOLOV4模型的果园樱桃实时检测研究[J]. 浙江农业学报, 2022, 34(11): 2522-2532.
[6]	张宁, 吴华瑞, 韩笑, 缪祎晟. 基于多尺度和注意力机制的番茄病害识别方法[J]. 浙江农业学报, 2021, 33(7): 1329-1338.