[1]谢瑞平,张志安.基于改进YOLOv5的复杂果园环境下梨桃果实轻量化检测与成熟度识别方法[J].机械与电子,2026,44(09):26-33.
　XIE Ruiping,ZHANG Zhian.Lightweight Detection and Maturity Recognition of Pear and Peach Fruits in Complex Orchard Environments Based on an Improved YOLOv5 Model[J].Machinery & Electronics,2026,44(09):26-33.
点击复制

基于改进YOLOv5的复杂果园环境下梨桃果实轻量化检测与成熟度识别方法()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
44
期数:
2026年09期
页码:
26-33
栏目:
智能检测
出版日期:
2026-09-25

文章信息/Info

Title:
Lightweight Detection and Maturity Recognition of Pear and Peach Fruits in Complex Orchard Environments Based on an Improved YOLOv5 Model
文章编号:
1001-2257(2026)09-0026-08
作者:
谢瑞平; 张志安
南京理工大学机械工程学院,江苏 南京 210094
Author(s):
XIE Ruiping; ZHANG Zhi’an
(School of Mechanical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)
关键词:
果实检测; 成熟度识别; YOLOv5; 轻量化
Keywords:
fruit detection; maturity recognition; YOLOv5; lightweight
分类号:
TP391.4
文献标志码:
A
摘要:
针对复杂果园环境下梨、桃果实检测精度不足以及现有模型计算开销较大的问题,提出一种基于改进YOLOv5的轻量化球类果实检测方法,实现梨、桃果实检测及成熟度识别。在模型设计上,引入 BiFPN 双向特征融合结构增强多尺度特征表达;设计PGR-C3轻量化模块,在降低模型计算量和参数量的同时保持特征提取能力;设计PR-ECA 注意力模块增强关键特征表征;采用WIoU 损失函数优化边界框回归精度。实验结果表明:改进模型的精确率、召回率和平均精度(mAP50)分别达到86.5%、84.0% 和89.2%,较原始YOLOv5分别提高4.9百分点、3.2百分点和2.3百分点;计算量、参数量和模型体积分别降低26.6%、25.9%和25.0%。与Faster R-CNN、SSD、YOLOv3、YOLOv8和YOLOv11等主流目标检测模型相比,所提方法在检测精度和轻量化性能方面均具有较好的综合优势。
Abstract:
To address the issues of insufficient detection accuracy for pear and peach fruits in complex orchard environments and the high computational cost of existing models,a lightweight spherical fruit detection method based on an improved YOLOv5 model is proposed for the simultaneous detection and maturity recognition.In model design,a BiFPN bidirectional feature fusion structure is introduced to enhance multi-scale feature representation;a PGR-C3 lightweight module is designed to reduce computational load and the number of parameters while preserving feature extraction capability;a PR-ECA attention module is developed to reinforce the representation of key features;and the WIoU loss function is adopted to optimize bounding box regression accuracy.Experimental results demonstrate that the improved model achieves a precision of 86.5%,recall of 84.0%,and a mean average precision (mAP50) of 89.2%,which are 4.9 percentage points,3.2 percentage points,and 2.3 percentage points higher than those of the original YOLOv5,respectively.Meanwhile,the computational complexity,the number of parameters,and the model size are reduced by 26.6%,25.9%,and 25.0%,respectively.Compared with mainstream object detection models such as Faster R-CNN,SSD,YOLOv3,YOLOv8,and YOLOv11,the proposed method demonstrates favorable comprehensive advantages in both detection accuracy and lightweight performance.

参考文献/References:

[1] 陈明猷,罗陆锋,刘威,等.采摘机器人全果园视觉感知及自主作业综述[J].智慧农业(中英文),2024,6(5):20-39.

[2] 周飞燕,金林鹏,董军.卷积神经网络研究综述[J].计算机学报,2017,40(6):1229-1251.
[3] 金保华,殷长魁,张卫正,等.基于机器视觉的苹果园果实识别研究综述[J].轻工学报,2019,34(2):71-81.
[4] Patel H N,Jain P K,Joshi M V.Fruit detection using improved multiple features based algorithm[J].International Journal of Computer Applications,2011,13(2):1-5.
[5] Yousuke K,Ninomiya Y,Kimura Y,et al.Strawberry cultivar identification and quality evaluation on the basis of multiple fruit appearance features[J].Computers and Electronics in Agriculture,2015,110:233-240.
[6] Ren Shaoqing,He Kaiming,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.
[7] Liu Wei,Anguelov D,Erhan D,et al.SSD:single shot multibox detector[C]∥Computer Vision – ECCV2016,2016:21-37.
[8] Redmon J,Farhadi A.YOLO9000:better,faster,stronger[ C]∥ 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).New York:IEEE,2017:7263-7271.
[9] Yan Bin,Fan Pan,Lei Xiaoyan,et al.A real-time apple targets detection method for picking robot based on improved YOLOv5[J].Remote Sensing,2021,13(9):1619.
[10] 刘天真,滕桂法,苑迎春,等.基于改进 YOLOv3 的自然场景下冬枣果实识别方法[J].农业机械学报,2021,52(5):17-25.
[11] Piotr D,Ron A,Serge B,et al.Fast feature pyramidsfor object detection[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2014,36(8):1532-1545.
[12] Liu Shu,Qi Lu,Qin Haifang,et al.Path aggregation network for instance segmentation[C]∥ 2018 IEEE/ CVF Conference on Computer Vision and Pattern Recognition (CVPR).New York:IEEE,2018:8759-8768.
[13] Tan Mingxing,Pang Ruoming,Le V Q,et al.Efficient Det:scalable and efficient object detection[C]∥ 2020 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).New York:IEEE,2020:10781-10790.
[14] Han Kai,Wang Yunhe,Tian Qi,et al.GhostNet:more features from cheap operations[C]∥ 2020 IEEE Computer Society Conference on Computer Visionand Pattern Recognition (CVPR).New York:IEEE,2020:1577-1586.
[15] Chen Jierun,Kao Shiuhong,He Hao,et al.Run,don’t walk:chasing higher FLOPs for faster neural networks[ C]∥ 2023 IEEE/CVF Conference on ComputerVision and Pattern Recognition (CVPR).New York::IEEE,2023:12021-12031.
[16] Wang Qilong,WuBanggu,Zhu Pengfei,et al.ECA-Net:efficient channel attention for deep convolutional neural net-works[C]∥ 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).New York:IEEE,2020:11531-11539.
[17] Zheng Zhaohui,Wang Ping,Ren Dongwei,et al.Enhancing geometric factors in model learning and inference for object detection and instance segmentation [J].IEEE Transactions on Cybernetics,2022,52(8):8574-8586.
[18] Tong Zanjia,Chen Yuhang,Xu Zewei,et al.Wise-IoU: bounding box regression loss with dynamic focusing mechanism[PP/OL].V3.arXiv (2023-04-08)[2026-05-10].https:∥arxiv.org/abs/2301.10051.

备注/Memo

备注/Memo:
收稿日期:2026-07-06
基金项目:江苏省现代农机装备与技术示范推广项目(NJ2023-13)
作者简介:谢瑞平 (2000-),男,江西赣州人,硕士研究生,研究方向为深度学习、机器人控制技术;张志安 (1979-),男,黑龙江绥化人,博士,副教授,研究方向为飞行器控制、智能控制算法和多编队机器人控制技术。
更新日期/Last Update: 2026-10-12