基于YOLOv4网络的棉花顶芽精准识别方法

doi:10.13304/j.nykjdb.2021.0536

中国农业科技导报 ›› 2022, Vol. 24 ›› Issue (8): 99-108.DOI: 10.13304/j.nykjdb.2021.0536

基于YOLOv4网络的棉花顶芽精准识别方法

刘海涛¹(), 韩鑫¹(), 兰玉彬¹^,², 伊丽丽¹, 王宝聚¹, 崔立华³

^1.山东理工大学农业工程与食品科学学院，山东淄博 255000
^2.山东省农业航空智能装备工程技术研究中心，山东淄博 255000
^3.山东绿风农业集团有限公司，山东滨州 256600

收稿日期:2021-07-02 接受日期:2021-09-22 出版日期:2022-08-15 发布日期:2022-08-22
通讯作者: 韩鑫
作者简介:刘海涛 E-mail：1243051772@qq.com；
基金资助:
财政部和农业农村部国家现代农业产业技术体系棉花产业体系项目(CARS-15-22);山东省引进顶尖人才“一事一议”专项(鲁政办字〔2018〕27号);淄博市重点研发计划项目(2019ZBXC053)

Precise Recognition Method of Cotton Top Buds Based on YOLOv4 Network

Haitao LIU¹(), Xin HAN¹(), Yubin LAN¹^,², Lili YI¹, Baoju WANG¹, Lihua CUI³

^1.School of Agricultural Engineering and Food Science，Shandong University of Technology，Shandong Zibo 255000，China
^2.Shandong Agricultural Aviation Intelligent Equipment Engineering Technology Research Center，Shandong Zibo 255000，China
^3.Shandong Lyufeng Agricultural Group Co. ，Ltd. ，Shandong Binzhou 256600，China

Received:2021-07-02 Accepted:2021-09-22 Online:2022-08-15 Published:2022-08-22
Contact: Xin HAN

摘要/Abstract

摘要：

为实现非接触、低成本、精准识别棉花顶芽，提出一种基于YOLOv4网络在复杂环境下对棉花顶芽进行精准识别的方法。利用K-means算法对棉花顶芽数据集进行聚类，优化先验框改善网络检测精度和速度，得到最优权值模型。对聚类前后模型以及与其他目标检测模型在棉花顶芽检测性能上进行了对比试验，并探究了顶芽在逆光和遮挡环境下，不同模型的检测性能。结果表明：该模型在测试集的平均检测精度（AP）、精确率（P）、召回率（R）、调和平均值（F₁）比原模型分别提高0.36%、1.73%、0.52%、1.16%，单张图像平均检测时间缩短0.28 s；对比SSD、YOLOv3、Tiny-YOLOV4模型，该模型检测精确率和F₁值最高，性能均衡；在自然场景处于逆光状态下，YOLOv4模型检测顶芽效果好于其他模型，且逆光环境对检测影响小；在遮挡条件下各个模型检测精度均有不同程度下降。

关键词: 棉花顶芽, YOLOv4, 深度学习, K-means, 图像识别

Abstract:

In order to achieve non-contact， low-cost accurate detection of cotton top buds， this paper proposed a recognition method based on the YOLOv4 network to detect the cotton top buds in a complex environment. The K-means algorithm was used to cluster the cotton top buds data set， and the anchors was optimized to improve the precision and speed of network detection， and the optimal model was selected through analysis. Compared with the performance of model before and after optimization， and the optimal weight model was compared with other object detection models in the detection performance of cotton top buds. The results showed that the average precision （AP） of optimal model in the test set， precision （P）， recall （R）， F₁ value of the optimize model increased 0.36%， 1.73%， 0.52%， 1.16% compared with the original YOLOv4 model， and the average detection time of a single image shortened 0.28 s； compared with SSD， YOLOv3， Tiny-YOLOV4 models， the optimize model had the highest F₁ value and recognition precision， and its performance was more balanced. The top buds at a backlight state in natural scenes had no effect on model detection， and the detection accuracy of each model decreased at different degrees under occlusion conditions.

Key words: cotton topping, YOLOv4, deep learning, K-means, image recognition

中图分类号:

S126

刘海涛, 韩鑫, 兰玉彬, 伊丽丽, 王宝聚, 崔立华. 基于YOLOv4网络的棉花顶芽精准识别方法[J]. 中国农业科技导报, 2022, 24(8): 99-108.

Haitao LIU, Xin HAN, Yubin LAN, Lili YI, Baoju WANG, Lihua CUI. Precise Recognition Method of Cotton Top Buds Based on YOLOv4 Network[J]. Journal of Agricultural Science and Technology, 2022, 24(8): 99-108.

图/表 12

参考文献 31

1	喻树迅.我国棉花生产现状与发展趋势[J].中国工程科学,2013,15(4):9-13.
	YU S X. Present situation and development trend of cotton production in China [J]. Strategic Study CAE, 2013,15(4):9-13.
2	卢秀茹,贾肖月,牛佳慧.中国棉花产业发展现状及展望[J].中国农业科学,2018,51(1):26-36.
	LU X R, JIA X Y, NIU J H. The present situation and prospects of cotton industry development in China [J]. Sci. Agric. Sin., 2018,51(1):26-36.
3	KONG H, YI L, LAN Y, et al.. Exploring the operation mode of spraying cotton defoliation agent by plant protection UAV [J]. Int. J. Precision Agric. Aviation, 2018, 1(1):19-24.
4	HAN X, YU J, LAN Y, et al.. Determination of application parameters for cotton defoliants in the Yellow River Basin [J]. Int. J. Precision Agric. Aviation, 2019, 2(1): 51-55.
5	YANG H, HU X, ZHAO J, et al.. Feature extraction of cotton plant height based on DSM difference method [J]. Int. J. Precision Agric. Aviation, 2018, 1(1):59-69.
6	牛巧鱼.我国棉花机械打顶研究进展[J].中国棉花,2013,40(11):23-24.
	NIU Q Y. Research progress of cotton topping machinery in China [J]. China Cott., 2013,40(11):23-24.
7	周桂鹏,张晓辉,范国强,等.棉花打顶机械化的研究现状及发展趋势[J].农机化研究,2014,36(4):242-245.
	ZHOU G P, ZHANG X H, FANG G Q, et al.. The research status and development trend on the cotton top-cutting mechanization [J]. J. Agric. Mechan. Res., 2014,36(4):242-245.
8	滕华灯,王春耀,蒋永新,等.两种棉花打顶机单体仿形机构的对比分析[J].西北农林科技大学学报(自然科学版),2011,39(3):163-167.
	TENG H D, WANG C Y, JIANG Y X, et al.. Contrast and analysis of the profile modeling mechanisms of two kinds of cotton topping monomers [J]. J. Northwest A&F Univ. (Nat. Sci. ), 2011,39(3):163-167.
9	姚强强,黄勇,陈永,等.单体仿形棉花打顶机的设计与试验[J].甘肃农业大学学报,2018,53(1):174-180, 186.
	YAO Q Q, HUANG Y, CHEN Y, et al.. Design and experimental of single profiling the cotton topping machine [J]. J. Gansu Agric. Univ., 2018,53(1):174-180, 186.
10	周海燕,尹素珍,朱立成,等.3WDZ-6型自走式棉花打顶机设计[J].农业机械学报,2010,41(S1):86-89.
	ZHOU H Y, YIN S Z, ZHU L C, et al.. Design of 3WDZ-6 self-propelled cotton top cutting [J]. Trans. Chin. Soc. Agric. Mach., 2010,41(S1):86-89.
11	彭强吉,荐世春,宋和平,等.3MDZJ-1型电力驱动式棉花智能精准打顶机的研制[J].农机化研究,2016,38(12):117-121.
	PENG Q J, JIAN S C, SONG H P, et al.. Development of 3MDZJ-1 type power driven intelligent precision cotton topping machine [J]. J. Agric. Mechan. Res., 2016,38(12):117-121.
12	宋欢,王维新,李霞,等.棉花打顶机械手的Fuzzy-PID控制[J].江苏农业科学,2017,45(3):172-175.
13	郝延杰,石磊,曹龙龙,等.棉花顶芽识别定位技术研究现状及展望[J].中国农机化学报,2018,39(11):72-78.
	HAO Y J, SHI L, CAO L L, et al.. Research status and prospect of cotton terminal bud identification and location technology [J]. J. Chin. Agric. Mechan., 2018,39(11):72-78.
14	刘俊奇. 棉花株顶识别系统的研究[D].石河子:石河子大学,2009.
	LIU J Q. The research of automatic recognition of cotton’s top [D]. Shihezi: Shihezi University, 2009.
15	瞿端阳.基于机器视觉技术的棉株识别系统研究[D].石河子:石河子大学,2013.
	QU D Y. Cotton plant recognition system based on the machine vision technology [D]. Shihezi: Shihezi University, 2013.
16	沈晓晨. 棉花打顶机棉株高度识别技术的研究[D].石河子:石河子大学,2018.
	SHEN X C. The height of the cotton plant identification technology for cotton top-cutting machine [D]. Shihezi: Shihezi University, 2018.
17	刘洋,战荫伟.基于深度学习的小目标检测算法综述[J].计算机工程与应用,2021,57(2):37-48.
	LIU Y, ZHAN Y W. Survey of small object detection algorithms based on deep learning [J]. Comput. Eng. Appl., 2021,57(2):37-48.
18	方路平,何杭江,周国民.目标检测算法研究综述[J].计算机工程与应用,2018,54(13):11-18, 33.
	FANG L P, HE H J, ZHOU G M. Research overview of object detection methods [J]. Comput. Eng. Appl., 2018,54(13):11-18, 33.
19	KRIZHEVSKY A, SUTSKEVER I, E.HINTON G. Image net classification with deep convolutional neural networks [J]. Adv. Neural Inform. Proc. Syst., 2012, 25(2)??.
20	BOCHKOVSKIY A, WANG C Y, LIAO H. YOLOv4: optimal speed and accuracy of object detection [J/OL]. 2020,10934 [2022-05-25]. .
21	刘小刚,范诚,李加念,等.基于卷积神经网络的草莓识别方法[J].农业机械学报,2020,51(2):237-244.
	LIU X G, FANG C, LI J N, et al.. Identification method of strawberry based on convolutional neural network [J]. Trans. Chin. Soc. Agric. Machinery, 2020,51(2):237-244.
22	吕石磊, 卢思华, 李震,等. 基于改进YOLOv3-LITE轻量级神经网络的柑橘识别方法[J]. 农业工程学报, 2019, 35(17):205-214.
	LYU S L, LU S H, LI Z, et al.. Orange recognition method using improved YOLOv3-LIT lightweight neural network [J]. Trans. Chin. Soc. Agric. Eng., 2019, 35(17):205-214.
23	赵德安, 吴任迪, 刘晓洋,等. 基于YOLO深度卷积神经网络的复杂背景下机器人采摘苹果定位[J]. 农业工程学报, 2019, 35(3):172-181.
	ZHAO D A, WU R D, LIU X Y, et al.. Apple positioning based on YOLO deep convolutional neural network for picking robot in complex background [J]. Trans. Chin. Soc. Agric. Eng., 2019, 35(3):172-181.
24	武星,齐泽宇,王龙军,等.基于轻量化YOLOv3卷积神经网络的苹果检测方法[J].农业机械学报,2020,51(8):17-25.
	WU X, QI Z Y, WANG L J, et al.. Apple detection method based on light-YOLOv3 convolutional neural network [J]. Trans. Chin. Soc. Agric. Mach., 2020,51(8):17-25.
25	燕红文,刘振宇,崔清亮,等.基于特征金字塔注意力与深度卷积网络的多目标生猪检测[J].农业工程学报,2020,36(11):193-202.
	YAN H W, LIU Z Y, CUI Q L, et al.. Multi-target detection based on feature pyramid attention and deep convolution network for pigs [J]. Trans. Chin. Soc. Agric. Eng., 2020,36(11):193-202.
26	HE K, ZHANG X, REN S, et al.. Spatial pyramid pooling in deep convolutional networks for visual recognition [J]. IEEE Trans. Pattern Anal. Mach. Int., 2015, 37(9):1904-1916.
27	LIU S, QI L, QIN H, et al.. Path aggregation network for instance segmentation [C]// Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2018.
28	ZHENG Z, WANG P, LIU W, et al.. Distance-iou loss: faster and better learning for bounding box regression [C]// Proceedings of AAAI Conference on Artificial Intelligence. AAAI, 2020.
29	李菊霞,李艳文,牛帆,等.基于YOLOv4的猪只饮食行为检测方法[J].农业机械学报,2021,52(3):251-256.
	LI J X, LI Y W, NIU F, et al.. Pig diet behavior detection method based on YOLOv4 [J]. Transact. Chin. Soc. Agric. Mach., 2021,52(3):251-256.
30	REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]// Proceedings of IEEE Conference on Computer Vision & Pattern Recognition. IEEE, 2017:6517-6525.
31	REDMON J, DIVVALA S, GIRSHICK R, et al.. You only look once: unified, real-time object detection [J/OL]. 2016,02640[2022-05-25]. .

光照 Light	植株状态 Plant status	图像数量 Number of images
晴天 Sunny	单株 Single plant	1 538
	多株 Multiple plants	429
	遮挡 Shelter	170
阴天 Cloudy	单株 Single plant	318
	多株 Multiple plants	24
	遮挡 Shelter	67
雨天 Rainy	单株 Single plant	254
	多株 Multiple plants	25
	遮挡 Shelter	75
总计Total		2 900

光照 Light	植株状态 Plant status	图像数量 Number of images
晴天 Sunny	单株 Single plant	1 538
	多株 Multiple plants	429
	遮挡 Shelter	170
阴天 Cloudy	单株 Single plant	318
	多株 Multiple plants	24
	遮挡 Shelter	67
雨天 Rainy	单株 Single plant	254
	多株 Multiple plants	25
	遮挡 Shelter	75
总计Total		2 900

模型Model	精确率 Precision/%	召回率 Recall/%	调和平均值 F₁/%	平均精度 Average precision/%
1	87.82	99.30	93.21	98.86
2	92.52	99.43	95.85	99.23
3	92.14	99.36	95.61	99.16
4	94.21	99.62	96.84	99.40
5	92.15	99.55	95.71	99.38
6	90.17	99.36	94.54	99.15
7	91.75	99.24	95.35	99.05
8	90.92	99.62	95.07	99.44
9	92.73	99.17	95.84	98.90

模型Model	精确率 Precision/%	召回率 Recall/%	调和平均值 F₁/%	平均精度 Average precision/%
1	87.82	99.30	93.21	98.86
2	92.52	99.43	95.85	99.23
3	92.14	99.36	95.61	99.16
4	94.21	99.62	96.84	99.40
5	92.15	99.55	95.71	99.38
6	90.17	99.36	94.54	99.15
7	91.75	99.24	95.35	99.05
8	90.92	99.62	95.07	99.44
9	92.73	99.17	95.84	98.90

模型 Model	精确率 Precision/%	召回率 Recall/%	调和平均值 F₁/%	平均精度 Average precision/%	检测时间 Time/s
YOLOv4	94.21	99.62	96.84	99.40	0.65
YOLOv3	93.73	99.23	96.40	99.17	0.84
Tiny-YOLOv4	72.56	99.70	83.99	99.48	0.53
SSD	72.03	99.80	83.67	99.51	0.78

基于YOLOv4网络的棉花顶芽精准识别方法

Precise Recognition Method of Cotton Top Buds Based on YOLOv4 Network

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 31

相关文章 10

编辑推荐

Metrics

模型 Model	逆光状态 Backlight environment				遮挡状态 Shelter environment
模型 Model	精确率 Precision	召回率 Recall	调和平均值 F₁	平均精度 Average precision	精确率 Precision	召回率 Recall	调和平均值 F₁	平均精度 Average precision
YOLOv3	92.82	99.48	96.03	99.41	92.31	99.31	95.68	98.93
YOLOv4	90.28	100.00	94.89	99.91	89.27	99.31	94.02	99.18
Tiny-YOLOv4	72.89	99.74	84.23	99.62	75.79	99.31	85.97	98.49
SSD	68.30	100.00	81.16	99.92	79.67	100.00	88.68	99.22

[1]	吴建伟, 黄杰, 熊晓菲, 高晗, 秦向阳. 基于AI的桃树病害智能识别方法研究与应用[J]. 中国农业科技导报, 2022, 24(5): 111-118.
[2]	高姻燕, 孙义, 李葆春. 基于无人机RGB影像估测田间小麦穗数[J]. 中国农业科技导报, 2022, 24(3): 103-110.
[3]	周惠汝, 吴波明. 深度学习在作物病害图像识别方面应用的研究进展[J]. 中国农业科技导报, 2021, 23(5): 61-68.
[4]	肖菲菲, 蒋蘋, 胡文武, 廖荣华, 张丹慧, 金生. 基于偏移补偿模型的极值点聚类苗带识别算法研究[J]. 中国农业科技导报, 2021, 23(5): 86-97.
[5]	吕纯阳, 刘升平, 郭秀明, 肖顺夫, 刘大众, 杨菲菲, 李路华. 基于SSD模型的巢门蜜蜂检测[J]. 中国农业科技导报, 2021, 23(5): 98-107.
[6]	许贝贝1,王文生1,2*,郭雷风1,陈桂鹏3. 基于非接触式的牛只身份识别研究进展与展望[J]. 中国农业科技导报, 2020, 22(7): 79-89.
[7]	刘哲,乌伟,张善文,崔倩倩,李瑞洋. 基于多元岭回归估计小麦种植密度[J]. 中国农业科技导报, 2020, 22(6): 72-80.
[8]	董伟1,钱蓉1,张洁2*,张立平1,陈红波2,张萌1,朱静波1,卜英乔3. 基于深度学习的蔬菜鳞翅目害虫自动识别与检测计数[J]. 中国农业科技导报, 2019, 21(12): 76-84.
[9]	夏雪1,周国民1*,丘耘1,李壮2,王健1,樊景超1,郭秀明1. 自然环境下苹果作业机器人双果重叠目标侦测方法[J]. 中国农业科技导报, 2018, 20(7): 63-73.
[10]	苏一峰§,杜克明§,李颖,孙忠富*,郑飞翔. 基于物联网平台的小麦病虫害诊断系统设计初探[J]. 中国农业科技导报, 2016, 18(2): 86-94.