[1]李颀,杨军.基于多分辨率特征融合的葡萄尺寸检测[J].江苏农业学报,2022,38(02):394-402.[doi:doi:10.3969/j.issn.1000-4440.2022.02.013]
　LI Qi,YANG Jun.Grape size detection based on multi-resolution feature fusion[J].,2022,38(02):394-402.[doi:doi:10.3969/j.issn.1000-4440.2022.02.013]
点击复制

基于多分辨率特征融合的葡萄尺寸检测()
分享到：

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
38
期数:
2022年02期
页码:
394-402
栏目:
农业信息工程
出版日期:
2022-04-30

文章信息/Info

Title:
Grape size detection based on multi-resolution feature fusion
作者:
李颀1; 杨军2
(1.陕西科技大学电子信息与人工智能学院，陕西西安710021；2.陕西科技大学电气与控制工程学院，陕西西安710021)
Author(s):
LI Qi1; YANG Jun2
(1.School of Electronic Information and Artificial Intelligence, Shaanxi University of Science and Technology, Xi′an 710021, China;2.School of Electrical and Control Engineering, Shaanxi University of Science and Technology, Xi′an 710021, China)
关键词:
葡萄无损检测; 多分辨率特征融合; 遮挡补偿; 机器视觉
Keywords:
nondestructive testing of grapes; multi-resolution feature fusion; occlusion compensation; machine vision
分类号:
TP391
DOI:
doi:10.3969/j.issn.1000-4440.2022.02.013
文献标志码:
A
摘要:
针对葡萄特征提取不够充分且果粒排列密集相互遮挡难以准确检测的问题，以陕西省鄠邑区户太8号葡萄为研究对象，提出一种基于特征金字塔网络(FPN)特征融合的Faster R-CNN卷积神经网络模型完成复杂背景情况下葡萄果粒的检测与识别。以ResNet50为主干网络，引入金字塔结构，增强网络模型对葡萄果粒不同分辨率特征的提取能力，同时加入GA-RPN网络生成自适应锚框，引入遮挡补偿机制，以解决密集葡萄果粒存在的遮挡问题。模型验证结果表明，本研究提出的模型精度均值（AP）在候选框与原标记框的重叠率（IOU）阈值为50时可达95.9%，对葡萄果粒、果穗的检测准确率分别为95.8%、96.1%，相比于原始Faster R-CNN模型识别性能更优。利用双目视觉算法对葡萄果粒进行尺寸测量，在最佳测量距离（0.6～1.4 m）其相对误差可控制在2%以内。
Abstract:
In view of the insufficient feature extraction of grapes and the difficulty in accurate detection of dense occlusion of grape grains, a Faster R-CNN convolutional neural network model based on feature pyramid network（FPN）feature fusion was proposed to detect and identify grape grains under complex background in Huyi District of Shaanxi Province. The ResNet50 was used as the main network, and the pyramid structure was introduced to enhance the extraction ability of the network model for different resolution characteristics of grape fruit. At the same time, the GA-RPN network was added to generate the adaptive anchor frame, and the occlusion compensation mechanism was introduced to solve the occlusion problem of dense grape fruit. The model verification results showed that the average precision (AP) of the proposed model could reach 95.9% when the threshold of overlap rate between candidate box and original marked box (IOU) was 50, and the detection accuracy of grapes and strings was 95.8% and 96.1%, respectively, which was better than that of the original Faster R-CNN model. Using binocular vision algorithm to measure the size of grape seeds, the relative error can be controlled within 2% at the optimal measurement distance (0.6-1.4 m).

参考文献/References:

[1]张帅亮. 作物生长过程检测技术研究与实现[D].新乡:河南科技学院,2018.
[2]杨涛,李晓晓. 机器视觉技术在现代农业生产中的研究进展[J].中国农机化学报,2021,42(3):171-181.
[3]ZHAO Y S,GONG L,HUANG Y X,et al. A review of key techniques of vision-based control for harvesting robot[J]. Computers and Electronics in Agriculture, 2016,127(6):312-321.
[4]XIONG J T,LIN R,LIU Z,et al. The recognition of litchi clusters and the calculation of picking point in a nocturnal natural environment[J]. Biosystems Engineering, 2018,166(8):44-57.
[5]李诗禹. 温棚环境及植物生长信息管理分析系统研究与实现[D].银川:宁夏大学,2018.
[6]赵献立,王志明. 机器学习算法在农业机器视觉系统中的应用[J].江苏农业科学,2020,48(12):226-231.
[7]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.
[8]叶浩,詹国敏,吴亮,等. 基于图像分割的葡萄果粒粒径快速测量技术[J].新技术新工艺,2016(5):84-87.
[9]罗陆锋,邹湘军,王成琳,等. 基于轮廓分析的双串叠贴葡萄目标识别方法[J].农业机械学报,2017,48(6):15-22.
[10]刘智杭. 基于图像的巨峰葡萄果粒尺寸检测方[D].哈尔滨:东北林业大学,2019.
[11]刘平,朱衍俊,张同勋,等. 自然环境下贴叠葡萄串的识别与图像分割算法[J].农业工程学报,2020,36(6):161-169.
[12]李睿. 基于机器视觉的苹果尺寸测量与分类研究[D].唐山:华北理工大学,2020.
[13]徐庆. 图像处理技术在现代农业中的应用[J].农村经济与科技,2020,31(1):66-67.
[14]QURESHI W S,PAYNE A,WALSH K B,et al. Machine vision for counting fruit on mango tree canopies[J]. Precision Agriculture, 2017,18(2):224-244.
[15]TANG Y C,CHEN M Y,WANG C L, et al. Recognition and localization methods for vision-based fruit picking robots: A review[J]. Frontiers in Plant Science,2020,5:1-17.
[16]李伟山,卫晨,王琳. 改进的Faster R-CNN煤矿井下行人检测算法[J].计算机工程与应用,2019,55(4):200-207.
[17]张守东,杨明,胡太. 基于多特征融合的显著性目标检测算法[J].计算机科学与探索,2019,13(5):834-845.
[18]吕晓琪,吴凉,谷宇等. 基于三维卷积神经网络的低剂量肺结节检测[J].光学精密工程,2018,26(5):1211-1218.
[19]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, 2017,39(6):1137-1149.
[20]SOHAIB A, MAHEDI H, SHEIKH M S I. YOLO-based three-stage network for Bangla license plate recognition in Dhaka Metropolitan City[J]. International Conference on Bangla Speech and Language Processing (ICBSLP), 2018,33(15):21-22.
[21]毋立芳,汪敏贵,付亨等. 深度目标检测与图像分类相结合的棉花发育期自动识别方法[J].中国科技论文,2018,13(20):2309-2316.
[22]张怡,赵珠蒙,王校常,等. 基于ResNet卷积神经网络的绿茶种类识别模型构建[J].茶叶科学,2021,41(2):261-271.
[23]王珂,王亮亮,李瑞峰,等. 基于折光鱼眼全景相机的双目立体视觉标定[J].中国科技论文,2016,11(8):913-919.
[24]卢艳军,王诗宇,张太宁,等. 针对目标跟踪过程的双目视觉测距方法研究[J].电光与控制,2020,27(1):60-63,101.
[25]MANSOOR A, MUHAMMAD S A, MUHAMMAD R, et al. Real-time machine-learning based Cropand weed detection and classification for variable-rate spraying in precision agriculture[J]. International Conference on Electrical and Electronics Engineering(ICEEE),2020,10(19):273-280.

备注/Memo

备注/Memo:
收稿日期：2021-07-06基金项目：陕西省农业科技创新工程资助项目
[201806117YF05NC13（1）]；陕西省科技厅农业科技攻关项目（2015NY028）；陕西科技大学博士科研启动基金资助项目（BJ13-15）作者简介：李颀（1973-），女，陕西西安人，博士，教授，主要研究方向为农业智能化信息化、深度学习。(E-mail)liqidq@sust.edu.cn通讯作者：杨军，(E-mail)1414831375@qq.com
更新日期/Last Update: 2022-05-07