Lightweight Grape Object Detection Fusion Algorithm Based on Improved YOLOv5s

doi:10.15933/j.cnki.1004-3268.2024.09.016

Abstract

Abstract: A lightweight grape target detection network YM‑GDM(YOLOv5s‑MobileNetV3 grape detection model) was proposed to meet the requirements of accuracy,real‑time performance,and lightweight of target detection models for agricultural automatic harvesting machinery.MobileNetV3 was employed as the backbone network instead of CSPDarknet53 in YOLOv5s to achieve model lightweighting.The introduction of Res2Net_C2f module and BiFPN(Bi‑directional feature pyramid network)structure aimed to enhance the model’s multi‑scale feature fusion capability.Additionally,the VariFocalLoss loss function was adopted to train the model,mitigating the impact of imbalanced positive and negative samples.The self‑made data set containing five types of table grapes and the open data set(WGISD)containing five types of wine grapes were used as test data sets.The experimental results showed that the YM‑GDM network achieved a mAP50 of 90. 8% for the detection of 10 grape classes.Compared to YOLOv3‑tiny,and YOLOv5s,it improved by 6.2,and 2.2 percentage points respectively.The model size was 9.73 MB,which was reduced by 44.4% and 32.8% compared to YOLOv3‑tiny and YOLOv5s,respectively.Furthermore,by further reducing the number of parameters,a lightweight specialized model,YM‑GDM‑tiny,was obtained with a model size of 4.73 MB and a mAP50 of 86.8%,suitable for deployment on mobile devices with lower computing power.

Key words: Grape, Object detection, YOLOv5s, Lghtweight network, Feature fusion

摘要： 针对农业自动采摘机械对目标检测模型准确率、实时性及轻量化的需求，提出了一种轻量级葡萄目标检测网络YM-GDM（YOLOv5s-MobileNetV3 grape detection model）。使用MobileNetV3 代替CSPDarknet53 作为YOLOv5s 的主干网络，以实现模型的轻量化；引入Res2Net_C2f 模块和BiFPN（Bi‑directional feature pyramid network）结构，以提高模型的多尺度特征融合能力；同时，改用VariFocalLoss损失函数对模型进行训练，以减少正负样本不均带来的影响。使用包含5类食用葡萄的自制数据集与包含5类酿酒葡萄的公开数据集（WGISD）作为试验数据集进行测试。结果表明，YMGDM网络对10个品种葡萄的目标检测平均精度均值（mAP50）达到90.8%，比YOLOv3-tiny、YOLOv5s分别提升6.2、2.2百分点；模型体积为9.72 MB，相比YOLOv3-tiny、YOLOv5s分别缩小44.4%、32.8%。此外，进一步减少参数量得到了轻量特化模型YM-GDM-tiny，模型体积缩小到4.73 MB，mAP50 达到86.8%，以部署于算力更低的移动设备。

关键词: 葡萄, 目标检测, YOLOv5s, 轻量化网络, 特征融合

CLC Number:

HU Junfeng, LI Songqing, HUANG Xiaowen, LIU Dayang, LI Baicong. Lightweight Grape Object Detection Fusion Algorithm Based on Improved YOLOv5s[J]. Journal of Henan Agricultural Sciences, 2024, 53(9): 150-158.

胡峻峰, 李松青, 黄晓文, 刘大洋, 李柏聪. 基于改进YOLOv5s 的轻量级葡萄目标检测融合算法[J]. 河南农业科学, 2024, 53(9): 150-158.

References

［1］蒙贺伟，李成松，坎杂，等.新疆酿酒葡萄振动采收装置设计与试验［J］.农机化研究，2017，39（10）：42‑45.

MENG H W，LI C S，KAN Z et al.The design and experiment of wine grape in Xinjiang vibration harvest device［J］.Journal of Agricultural Mechanization Research，2017，39（10）：42‑45.

［2］LUO L F，TANG Y C，ZOU X J，et al.Vision‑based extraction of spatial information in grape clusters for harvesting robots［J］.Biosystems Engineering，2016，151：90‑104.

［3］NUSKE S，ACHAR S，BATES T，et al.Yield estimation in vineyards by visual grape detection［C］//2011 IEEE/RSJ International Conference on Intelligent Robots and Systems. San Francisco，CA，USA：IEEE，2011：2352‑2358.

［4］LIU S，WHITTY M.Automatic grape bunch detection in vineyards with an SVM classifier［J］.Journal of Applied Logic，2015，13（4）：643‑653.

［5] BADEKA E，KALABOKAS T，TZIRIDIS K，et al.Grapes visual segmentation for harvesting robots using local texture descriptors［C］//International Conference on Computer Vision Systems. Cham：Springer International Publishing，2019：98‑109.

［6］陈英，李伟，张俊雄.基于图像轮廓分析的堆叠葡萄果粒尺寸检测［J］.农业机械学报，2011，42（8）：168‑172.

CHEN Y，LI W，ZHANG J X.Overlapped grapes berry size inspection based on image contour analysis［J］.Transactions of the Chinese Society for Agricultural Machinery，2011，42（8）：168‑172.

［7］FONT D，PALLEJÀ T，TRESANCHEZ M，et al.Counting red grapes in vineyards by detecting specular spherical reflection peaks in RGB images obtained at night with artificial illumination［J］.Computers and Electronics in Agriculture，2014，108：105‑111.

［8］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.

LUO L F，ZOU X J，WANG C L，et al.Recognition method for two overlapping and adjacent grape clusters based on image contour analysis［J］.Transactions of the Chinese Society for Agricultural Machinery，2017，48（6）：15‑22.

［9］田锐，郭艳玲.基于机器视觉的葡萄自动识别技术［J］.东北林业大学学报，2008，36（11）：95‑97.

TIAN R，GUO Y L.Automatic identification techniques of grape based on computer vision［J］.Journal of Northeast

Forestry University，2008，36（11）：95‑97.

［10］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.

LÜ S L，LU S H，LI Z，et al.Orange recognition method using improved YOLOv3‑LITE lightweight neural network［J］.Transactions of the Chinese Society of Agricultural Engineering，2019，35（17）：205‑214.

［11］TIAN Y N，YANG G D，WANG Z，et al.Apple detection during different growth stages in orchards using the improved YOLO‑V3 model［J］.Computers and Electronics in Agriculture，2019，157：417‑426.

［12］SU S，CHEN R，FANG X，et al.A novel lightweight grape detection method［J］.Agriculture，2022，12（9）：1364.

［13］李阳德，马晓慧，王骥.基于轻量级MobileNet V3-YOLOv4的生长期菠萝成熟度分析［J］.智慧农业（中英文），2023，5（2）：35‑44.

LI Y D，MA X H，WANG J.Pineapple maturity analysis in natural environment based on MobileNet V3-YOLOv4［J］.Smart Agriculture，2023，5（2）：35‑44.

［14］李惠鹏，李长勇，李贵宾，等.基于深度学习的多品种鲜食葡萄采摘点定位［J］.中国农机化学报，2022，43 （12）：155‑161.

LI H P，LI C Y，LI G B，et al.Picking point positioning of multi‑variety table grapes based on deep‑learning［J］.Journal of Chinese Agricultural Mechanization，2022，43（12）：155‑161.

［15］SANTOS T T，DE SOUZA L L，DOS SANTOS A A，et al.Grape detection，segmentation，and tracking using deep neural networks and three‑dimensional association［J］.Computers and Electronics in Agriculture，2020，170：105247.

［16］王梁，侯义锋，贺杰.基于Mask-RCNN的自然场景下油茶果目标识别与检测［J］.中国农机化学报，2022，43（12）：148‑154.

WANG L，HOU Y F，HE J.Target recognition and detection of Camellia oleifera fruit in natural scene based on Mask‑RCNN［J］.Journal of Chinese Agricultural Mechanization，2022，43（12）：148‑154.

［17］李伟豪，詹炜，周婉，等.轻量型Yolov7-TSA网络在茶叶病害检测识别中的研究与应用［J］.河南农业科学，2023，52（5）：162‑169.

LI W H，ZHAN W，ZHOU W，et al.Research and application of lightweight Yolov7‑TSA network in tea disease detection and identification［J］.Journal of Henan Agricultural Sciences，2023，52（5）：162‑169.

［18］LI H，LI C，LI G，et al.A real‑time table grape detection method based on improved YOLOv4‑tiny network in complex background［J］.Biosystems Engineering，2021，212：347‑359.

［19］李国进，黄晓洁，李修华，等.采用轻量级网络MobileNetV2的酿酒葡萄检测模型［J］.农业工程学报，2021，37（17）：168‑176.

LI G J，HUANG X J，LI X H，et al.Detection model for wine grapes using MobileNetV2 lightweight network［J］.Transactions of the Chinese Society of Agricultural Engineering，2021，37（17）：168‑176.

［20］HOWARD A，SANDLER M，CHU G，et al.Searching for mobilenetv3［C］//Proceedings of the IEEE/CVF International Conference on Computer Vision（ICCV）.Seoul，Korea（South）：IEEE，2019：1314‑1324.

［21］GAO S H，CHENG M M，ZHAO K，et al.Res2net：A new multi‑scale backbone architecture［J］.IEEE Transactions on Pattern Analysis and Machine Intelligence，2019，43（2）：652‑662.

［22］ZHANG H，WANG Y，DAYOUB F，et al.Varifocalnet：An IoU‑aware dense object detector［C］//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition（CVPR）.Nashville，TN，USA：IEEE，2021：8514‑8523.

[1]	DAI Jiabing, SONG Chunfang, LING Caijin, LI Zhenfeng, SUN Chonggao. Research on Insect‑bitten Zijin Tea Detection Method Based on YOLOv5s‑SE and Channel Pruning [J]. Journal of Henan Agricultural Sciences, 2024, 53(5): 157-163.
[2]	CUI Xiaoyue, SHANG Hongquan, LÜ Zhongwei, LOU Yusui, ZHANG Ke, FAN Hongjie, WU Wenying, ZHANG Xiaofeng. Cloning and Expression Analysis of NF‑YB3 Gene from Vitis vinifera [J]. Journal of Henan Agricultural Sciences, 2024, 53(4): 111-118.
[3]	LI Dongya, BAT Tao, XIANG Huimin, DAI Shuo, WANG Zhenlu, CHEN Zhen. Recognition of Cotton Pests and Diseases Named Entities Based on RoBERTA Multi⁃feature Fusion [J]. Journal of Henan Agricultural Sciences, 2024, 53(2): 152-161.
[4]	ZHONG Zhengyang, YUN Lijun, YANG Xuanxi, CHEN Zaiqing. Research on Walnut Recognition Algorithm in Natural Environment Based on Improved YOLOX [J]. Journal of Henan Agricultural Sciences, 2024, 53(1): 152-161.
[5]	ZHONG Bin, YANG Jun, LIU Yi, REN Jintao. Target Detection of Pasture Weeds Based on Improved DINO [J]. Journal of Henan Agricultural Sciences, 2023, 52(9): 156-163.
[6]	LOU Yusui, SHANG Hongquan, FAN Hongjie, LI Zheng, ZHANG Ke, CUI Xiaoyue, LÜ Zhongwei. Effects of Different Color Fruit Bags on Berry Russet and Quality of Shine Muscat Grape during Ripening [J]. Journal of Henan Agricultural Sciences, 2023, 52(8): 105-114.
[7]	WANG Huanxin, SHEN Zhihao, LIU Quan, LIU Jinjiang. Identification of Crop Leaf Diseases Based on Improved MobileNetV2 Model [J]. Journal of Henan Agricultural Sciences, 2023, 52(4): 143-151.
[8]	WEI Gaixing, YI Wenlong, LIU Yucheng, ZHAO Yingding, CHEN Tingzhuo. Research on Improved U‑Net Method for Rice Leaf Cell Segmentation [J]. Journal of Henan Agricultural Sciences, 2023, 52(3): 153-160.
[9]	LI Rihui, ZHONG Xiaomin, DUAN Xinyao, YAN Yasai, WANG Lin, WANG Hua, LI Hua. The Impact of Simplified Ecological Cultivation Model on the Soil Quality of Italian Riesling Vineyard [J]. Journal of Henan Agricultural Sciences, 2023, 52(12): 127-135.
[10]	WANG Yingchao, ZHANG Jingjing, JIA Donglin, ZHOU Tengfei. Grading of Apple Based on K‑means Clustering and Improved MLP [J]. Journal of Henan Agricultural Sciences, 2023, 52(1): 161-171.
[11]	LOU Yusui, SHANG Hongquan, FAN Hongjie, LI Zheng, ZHANG Ke, LÜ Zhongwei, ZHANG Xiaofeng, WU Wenying. Changes in Berry Peel Structure and Related Substances of Shine Muscat Grape with Different Berry Russet Levels [J]. Journal of Henan Agricultural Sciences, 2022, 51(9): 105-113.
[12]	LOU Yusui, ZHANG Xiaofeng, LI Zheng, SHANG Hongquan, FAN Hongjie, LÜ Zhongwei. Effect of Different Yield Load on Quality Development during Delayed Harvest of Shine Muscat Grape [J]. Journal of Henan Agricultural Sciences, 2022, 51(6): 111-118.
[13]	WANG Xingwang, ZHENG Hanyuan, WANG Suqing. Identification of Diseases and Insect Pests of Malu Grape Based on CSJMM⁃AS⁃GAC [J]. Journal of Henan Agricultural Sciences, 2022, 51(6): 154-163.
[14]	LIU Jiangpeng, NIU Qunfeng, JIN Yi, CHEN Xia, WANG Li, YUAN Qiang. Research on Tobacco Shred Image Recognition Method Based on Efficient Channel Attention Mechanism and Multi‑Scale Feature Fusion [J]. Journal of Henan Agricultural Sciences, 2022, 51(11): 145-154.
[15]	SONG Shen, SHAN Shouming, LI Qian, LIU Sifan. Effects of Sink-Source Relationship on Anthocyanin Accumulation and Quality of Beimei Grape [J]. Journal of Henan Agricultural Sciences, 2021, 50(6): 109-115.