基于改进YOLOv7的茶叶嫩芽识别模型研究

doi:10.12398/j.issn.2096-7217.2024.02.005

摘要/Abstract

摘要：

为了在复杂环境中有效识别茶叶嫩芽并在最大限度避免伤害茶树的情况下，提高智能化采摘精度，本研究针对传统目标检测算法在茶园中存在的检测精度低、鲁棒性差等问题，提出一种基于改进YOLOv7的茶叶嫩芽识别与检测的模型YOLOv7-tea，从而实现对茶叶嫩芽的快速识别与检测。首先，采集茶叶嫩芽图像并完成嫩芽图像的标注和数据增强，构建茶叶嫩芽数据集。其次，通过在YOLOv7主干网络的3个特征提取层中引入CBAM注意力机制模块，增强模型对特征提取的能力；采用SPD-Conv模块替换颈部网络下采样模块中的SConv模块，以减少小目标特征的丢失；通过EIoU损失函数优化框回归损失，提升预测框的准确率。最后，以茶叶嫩芽图像数据集为样本将其他目标检测模型与YOLOv7-tea模型进行对比试验，并对不同距离、不同角度拍摄的茶叶嫩芽图像进行识别效果检测。试验结果表明，YOLOv7-tea网络模型比YOLOv7模型的精确率（Precision， P）、召回率（Recall， R）和识别平均精度（mAP）值分别高出2.87、6.91和8.69个百分点，且模型的检测速度更快，在复杂背景下对茶叶嫩芽的识别与检测具有更高的置信度分数。该研究构建的YOLOv7-tea模型对小尺寸茶叶嫩芽的识别效果较好，减少了漏检和误检的情况，具有良好的鲁棒性和实时性，可为茶叶的产量预估和智能化采摘提供参考。

关键词: 茶叶嫩芽, 目标检测, CBAM注意力机制, 自动化采摘, YOLOv7

Abstract:

To effectively identify tea buds in complex environments and improve the precision of intelligent harvesting while minimizing damage to tea trees， this study addresses the issues of low detection accuracy and poor robustness exhibited by traditional target detection algorithms in tea gardens， and proposes YOLOv7-tea model for tea bud identification and detection based on an improved YOLOv7， so as to achieve rapid recognition and detection of tea buds.First， tea bud images were collected and annotated， and data augmentation was performed to construct a tea bud dataset. Next， the CBAM attention mechanism module was introduced into three feature extraction layers of the YOLOv7 backbone network to enhance the model's feature extraction capability； the SPD-Conv module was used to replace the SConv module in the neck network's downsampling module to reduce the loss of small object features； and the EIoU loss function was employed to optimize box regression， thereby improving the accuracy of the predicted boxes. Finally， a comparative experiment was conducted between other target detection models and the YOLOv7-tea model using the tea bud image dataset as a sample， and the recognition effect of tea buds shot at different distances and angles was tested.The experimental results show that the YOLOv7-tea network model outperforms the YOLOv7 model in terms of precision （P）， recall （R）， and mean average precision （mAP） by 2.87， 6.91， and 8.69 percentage points， respectively. Additionally， the model has a faster detection speed and exhibits higher confidence scores in the recognition and detection of tea buds in complex backgrounds.The YOLOv7-tea model constructed in this study demonstrates better recognition performance for small-sized tea leaf buds， reducing instances of missed detection and false alarms. It exhibits good robustness and real-time performance， offering valuable insights for estimating tea yield and implementing intelligent harvesting.

Key words: tea buds, object detection, CBAM attention mechanism, automated picking, YOLOv7

中图分类号:

魏堂伟, 张津诚, 王晶, 周庆燕. 基于改进YOLOv7的茶叶嫩芽识别模型研究[J]. 智能化农业装备学报（中英文）, 2024, 5(2): 42-50.

WEI Tangwei, ZHANG Jincheng, WANG Jing, ZHOU Qingyan. Study of tea buds recognition and detection based on improved YOLOv7 model[J]. Journal of Intelligent Agricultural Mechanization, 2024, 5(2): 42-50.

图/表 10

图1 不同角度和距离拍摄的茶叶嫩芽

Figure 1 Pictures of tea buds taken from different angles and distances

图2 茶叶嫩芽数据增强

Figure 2 Tea bud date enhancement

图3 CBAM注意力机制结构图

Figure 3 CBAM structure

图4 引入CBAM注意力模块后的YOLOv7网络结构图

Figure 4 YOLOv7 network structure diagram after introducing CBAM attention module

图5 当scale=2时SPD-Conv的图解

Figure 5 Illustration of SPD-Conv when scale=2

图6 改进后MP2结构图

Figure 6 Improved MP2 structure diagram

表1 消融试验结果

Table 1 Results of ablation test

模型	CBAM	SPD-Conv	EIoU_Loss	mAP/%	P/%	FPS/(帧·s^-1)
YOLOv7				71.48	87.45	56.6
YOLOv7-2	√			78.28	88.89	55.7
YOLOv7-3	√	√		79.95	90.27	55.4
YOLOv7-tea	√	√	√	80.17	90.32	57.2

图7 YOLOv7-tea loss曲线

Figure 7 YOLOv7-tea loss curve

表2 不同检测算法的试验结果对比

Table 2 Comparison of experimental results from different detection algorithms

模型	F1值	mAP /%	R /%	FPS /(帧·s^-1)	P /%
Faster R-CNN	0.68	51.94	63.52	21.3	72.12
YOLOv5	0.59	56.48	48.77	49.0	73.46
YOLOX	0.65	59.61	60.25	55.3	71.36
YOLOv7	0.76	71.48	66.77	56.6	87.45
YOLOv7-tea	0.81	80.17	73.68	57.2	90.32

图8 五种目标检测算法对茶叶嫩芽的检测效果

Figure 8 The detection effect of five target detection algorithms on tea buds

参考文献 21

1	江用文，陈霄雄，朱建淼，等. 中国茶产业2020年发展规模分析［J］. 茶叶科学， 2011， 31（3）： 273-282．
	JIANG Yongwen， CHEN Xiaoxiong， ZHU Jianmiao， et al. Analysis on development scale of Chinese tea industry in 2020 ［J］. Journal of Tea Science， 2011， 31（3）： 273-282.
2	TIAN L， HE Y. Study on the cross-cultural marketing strategies of Chinese tea enterprises ［J］. Academic Journal of Business and Management， 2022， 4（2）： 38-41.
3	李宗洙. 我国茶产业贸易发展研究：模式、问题及对策建议［J］. 商业经济研究， 2021（24）： 186-189．
4	韩余，宋志禹，陈巧敏. 4CJ-1200F智能采茶机设计与试验［J］. 智能化农业装备学报（中英文）， 2022， 3（1）： 1-6.
	HAN Yu， SONG Zhiyu， CHEN Qiaomin. Design and experiment of 4CJ-1200F intelligent tea plucking machine ［J］. Journal of Intelligent Agricultural Mechanization， 2022， 3（1）： 1-6.
5	张浩，陈勇，汪巍，等. 基于主动计算机视觉的茶叶采摘定位技术［J］. 农业机械学报， 2014， 45（9）： 61-65.
	ZHANG Hao， CHEN Yong， WANG Wei， et al. Positioning method for tea picking using active computer vision ［J］. Transactions of the Chinese Society for Agricultural Machinery， 2014， 45（9）： 61-65.
6	许宝阳，高延峰. 基于Faster-RCNN深度学习的茶叶嫩芽多维度识别及其性能分析［J］. 农业装备与车辆工程， 2023， 61（2）： 19-24.
	XU Baoyang， GAO Yanfeng. Tea-buds multi-dimensional recognition with Faster-RCNN deep learning method and its performance analysis ［J］. Agricultural Equipment & Vehicle Engineering， 2023， 61（2）： 19-24.
7	SHAO P D， WU M H， WANG X W， et al. Research on the tea bud recognition based on improved k-means algorithm ［C］// MATEC Web of Conferences. EDP Sciences， 2018， 232： 03050.
8	YAN C， CHEN Z， LI Z， et al. Tea sprout picking point identification based on improved DeepLabV3+［J］. Agriculture， 2022， 12（10）： 1594.
9	ZHANG F Y， SUN H W， XIE S， et al. A tea bud segmentation，detection and picking point localization based on the MDY7-3PTB model ［J］. Frontiers in Plant Science， 2023： 1199473-1199473.
10	CHENG Y F， LI Y， ZHANG R T. Locating tea bud keypoints by keypoint detection method based on convolutional neural network ［J］. Sustainability， 2023， 15（8）： 6898.
11	LI J， LI J H， ZHAO X， et al. Lightweight detection networks for tea bud on complex agricultural environment via improved YOLO v4 ［J］. Computers and Electronics in Agriculture， 2023， 211（C）： 107955.
12	龙樟，姜倩，王健，等. 茶叶嫩芽视觉识别与采摘点定位方法研究［J］. 传感器与微系统， 2022， 41（2）： 39-41， 45.
	LONG Zhang， JIANG Qian， WANG Jian， et al. Research on method of tea flushes vision recognition and picking point localization ［J］. Transducer and Microsystem Technologies， 2022， 41（2）： 39-41， 45.
13	杨福增，杨亮亮，田艳娜，等. 基于颜色和形状特征的茶叶嫩芽识别方法［J］. 农业机械学报， 2009， 40（S1）： 119-123.
	YANG Fuzeng， YANG Liangliang， TIAN Yanna， et al. Recognition of the tea sprout based on color and shape features ［J］. Transactions of the Chinese Society for Agricultural Machinery， 2009， 40（S1）： 119-123.
14	王梦妮，顾寄南，王化佳，等. 基于改进YOLOv5s模型的茶叶嫩芽识别方法［J］. 农业工程学报， 2023， 39（12）： 150-157.
	WANG Mengni， GU Ji'nan， WANG Huajia， et al. Method for identifying tea buds based on improved YOLOv5s model ［J］. Transactions of the Chinese Society of Agricultural Engineering， 2023， 39（12）： 150-157.
15	俞龙，黄楚斌，唐劲驰，等. 基于YOLOX改进模型的茶叶嫩芽识别方法［J］. 广东农业科学， 2022， 49（7）： 49-56.
	YU Long， HUANG Chubin， TANG Jinchi， et al. Tea bud recognition method based on improved YOLOX model ［J］. Guangdong Agricultural Sciences， 2022， 49（7）： 49-56.
16	WANG C Y， BOCHKOVSKIY A， LIAO H Y 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， 2023： 7464-7475.
17	郝紫霄，王琦. 基于YOLO-v7的无人机航拍图像小目标检测改进算法［J］. 软件导刊， 2024（1）： 167-172.
	HAO Zixiao， WANG Qi. Enhanced algorithm for small target detection in UAV aerial images based on YOLO-v7 ［J］. Software Guide， 2024（1）： 167-172.
18	WOO S， PARK J， LEE J Y， et al. CBAM： Convolutional block attention module ［C］// Proceedings of the European conference on computer vision （ECCV）. 2018： 3-19.
19	胡和平，吴明晖，洪孔林，等. 基于改进YOLOv5s的茶叶嫩芽分级识别方法［J］. 江西农业大学学报， 2023， 45（5）： 1261-1272.
	HU Heping， WU Minghui， HONG Konglin， et al. Classification and recognition method for tea buds based on improved YOLOv5s ［J］. Acta Agriculturae Universitatis Jiangxiensis， 2023， 45（5）： 1261-1272.
20	SUNKARA R， LUO T. No more strided convolutions or pooling： A new CNN building block for low-resolution images and small objects ［C］// Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Cham： Springer Nature Switzerland， 2022： 443-459.
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.

[1]	黄成龙, 张忠福, 卢智浩, 张晓君, 朱龙付, 杨万能. 基于VFNet-Improved和Deep Sort的棉花黄萎病病情分级[J]. 智能化农业装备学报（中英文）, 2023, 4(2): 12-21.
[2]	肖张娜, 罗陆锋, 陈明猷, 王金海, 卢清华, 骆少明. 基于改进YOLO-v4的果园环境下葡萄检测[J]. 智能化农业装备学报（中英文）, 2023, 4(2): 35-43.
[3]	李搴曦, 孙晓明, 江晗慧, 吴爱茹, 傅隆生, 李瑞. 基于YOLOv4-tiny的设施番茄智能喷药无人车设计与试验[J]. 智能化农业装备学报（中英文）, 2023, 4(2): 44-52.