[1]刘芳,刘玉坤,张白.基于D-YOLOv3检测网络的温室叶菜幼苗图像检测[J].江苏农业学报,2021,(05):1262-1269.[doi:doi:10.3969/j.issn.1000-4440.2021.05.022]
　LIU Fang,LIU Yu-kun,ZHANG Bai.Image detection of leafy vegetable seedlings in greenhouse based on D-YOLOv3 detection network[J].,2021,(05):1262-1269.[doi:doi:10.3969/j.issn.1000-4440.2021.05.022]
点击复制

基于D-YOLOv3检测网络的温室叶菜幼苗图像检测()
分享到：

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2021年05期
页码:
1262-1269
栏目:
园艺
出版日期:
2021-10-30

文章信息/Info

Title:
Image detection of leafy vegetable seedlings in greenhouse based on D-YOLOv3 detection network
作者:
刘芳; 刘玉坤; 张白
（北方民族大学电气信息工程学院，宁夏银川750021）
Author(s):
LIU Fang; LIU Yu-kun; ZHANG Bai
(College of Electrical Information Engineering, North Minzu University, Yinchuan 750021, China )
关键词:
小目标; 图像检测; 密集连接; D-YOLOv3检测网络
Keywords:
small target; image detection; dense connection; D-YOLOv3 detection network
分类号:
S636;TP391.41
DOI:
doi:10.3969/j.issn.1000-4440.2021.05.022
文献标志码:
A
摘要:
为解决温室叶菜子叶期幼苗生长密集情况下的图像识别问题，提出一种密集连接型D-YOLOv3检测网络。该网络以YOLOv3为基础构建主干网络,改进检测结构和损失函数。以穴盘培育的油菜幼苗为例展开一系列试验。首先确定了YOLOv3和D-YOLOv3检测网络中损失函数的修正系数；其次通过构建的几种检测网络的对比试验验证了对YOLOv3主干网络、检测结构和损失函数改进的有效性，D-YOLOv3的幼苗检测精度高达93.44%，检测时间低至12.61 ms，与YOLOv3相比精度提升9.4个百分点，时间降低4.07 ms；最后进行不同密集程度和光照环境下幼苗图像的检测性能对比试验，结果表明D-YOLOv3的检测精度、检测时间及对小目标的特征提取能力均优于YOLOv3。D-YOLOv3能够对温室环境下的叶菜幼苗进行有效检测，可以为智能检测装备的作物识别提供依据。
Abstract:
In order to solve the problem of detecting densely planted leafy vegetable seedlings at cotyledon stage in greenhouse, a densely connected detection network named D-YOLOv3 was proposed. The backbone network of D-YOLOv3 was built based on that of YOLOv3. Simultaneously, the detection structure and loss function of D-YOLOv3 were constructed as the improvement of those of YOLOv3. A series of comparative tests were carried out with rape seedlings cultivated in pot as an example. Firstly, the correction coefficients of loss function in YOLOv3 and D-YOLOv3 detection networks were determined. Then, the effectiveness of the derived backbone network, detection structure and loss function of D-YOLOv3 was verified according to comparative tests with some networks mentioned in the paper. The test results showed that the detection accuracy of D-YOLOv3 was as high as 93.44%, and the detection time was as low as 12.61 ms. Compared with YOLOv3, the detection accuracy was improved by 9.4 percentage point and the detection time was reduced by 4.07 ms. Finally, the performance of YOLOv3 and D-YOLOv3 detection networks was obtained with images of rape seedlings growing under different density and illumination environments. The detection precision and speed of D-YOLOv3 for the proposed image were better than those of YOLOv3 at various densities. The ability of D-YOLOv3 was stronger than that of YOLOv3 to extract features of small targets under different illumination and density conditions. Consequently, the D-YOLOv3 detection network proposed in this paper can effectively identify rape seedlings in the greenhouse and serve as a powerful tool for crop recognition with intelligent detection equipment.

参考文献/References:

[1]GRINBLAT G L, UZAL L C, LARESE M G, et al. Deep learning for plant identification using vein morphological patterns [J]. Computers and Electronics in Agriculture, 2016, 127: 418-424.
[2]DYRMANN M, KARSTOFT H, MIDTIBY H S. Plant species classification using deep convolutional neural network [J]. Biosystems Engineering, 2016, 151: 72-80.
[3]OPPENHEIM D, SHANI G. Potato disease classification using convolution neural networks [J]. Advances in Animal Biosciences, 2017, 8: 244-249.
[4]LU H, CAO Z, XIAO Y, et al. Tasselnet: Counting maize tassels in the wild via local counts regression network [J]. Plant Method, 2017, 13: 79-92.
[5]刘芳,刘玉坤,林森,等. 基于改进型YOLO的复杂环境下番茄果实快速识别方法[J].农业机械学报, 2020,51(6):229-237.
[6]TELLAECHE A, PAJARES G, BURGOS-ARTIZZU X P, et al. A computer vision approach for weeds identification through Support Vector Machines [J]. Applied Soft Computing, 2011, 11(1): 908-915.
[7]JAGAN K M, BALASUBRAMANIAN M, PALANIVEL S. Detection and recognition of diseases from paddy plant leaf images [J]. International Journal of Computer Applications, 2016, 144: 34-41.
[8]崔日鲜,刘亚东,付金东. 基于机器学习和可见光光谱的冬小麦叶片氮积累量估算[J].光谱学与光谱分析, 2016, 36(6): 1837-1842.
[9]DUAN L, HUANG C, CHEN G, et al. Determination of rice panicle numbers during heading by multi-angle imaging[J]. The Crop Journal, 2015, 3: 211-219.
[10]SADEGHI T P, SABERMANESH K, VIRLET N, et al. Automated method to determine two critical growth stages of wheat: Heading and flowering [J]. Front Plant Science, 2017, 8: 252-265.
[11]翁杨,曾睿,吴陈铭,等. 基于深度学习的农业植物表型研究综述[J]. 中国科学: 生命科学, 2019, 49(6): 698-716.
[12]GIRSHICK R. Fast R-CNN[C]. Santiago, Chile: Proceedings of the IEEE international conference on computer vision, 2015: 1440-1448.
[13]REN S, HE K, 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, 2017, 39(6):1137-1149.
[14]REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]. Ohio: conference on Computer Vision and Pattern Recognition, 2016: 779-788.
[15]REDMON J, FARHADI A. YOLO9000: better, faster,stronger[C]. Ohio: Conference on Computer Vision and Pattern Recognition, 2017: 6517-6525
[16]
[16] REDMON J, FARHADI A. YOLOv3: An incremental improvement [DB/OL]. (2018-04-18)
[2020-03-01]. https://arxiv.org/pdf/1804.02767.pdf.
[17]孙哲,张春龙,葛鲁镇,等. 基于Faster R-CNN的田间西兰花幼苗图像检测方法[J].农业机械学报, 2019, 50(7): 216-221.
[18]GAO H，ZHANG L．Densely connected convolutional networks[C]. Ohio: Conference on Computer Vision and Pattern Recognition, 2017:1-28．
[19]吴天舒,张志佳,刘云鹏,等. 基于改进SSD的轻量化小目标检测算法[J]. 红外与激光工程，2018，47(7)：47-53.
[20]KISANTAL M, WOJNA Z, MURAWSKI J, et al. Augmentation for small object detection[DB/OL]. (2019-02-19)
[2020-03-01]. https://www.researchgate.net/publication/338171640.
[21]冯小雨,梅卫,胡大帅. 基于改进Faster R-CNN的空中目标检测[J].光学学报,2018,38(6):250-258.

备注/Memo

备注/Memo:
收稿日期：2021-01-25基金项目：宁夏高等学校一流学科建设（电子科学与技术学科）项目（NXYLXK2017A07）；宁夏自然科学基金项目（2021AAC03229）；宁夏智慧农业产业技术协同创新中心基金项目作者简介：刘芳(1970-)，女，宁夏中卫人，博士，教授，博士生导师，主要从事智能控制研究，(E-mail)fangliu214@163.com
更新日期/Last Update: 2021-11-09