Detection of Dense Citrus Fruits by Combining Coordinated Attention and Cross-Scale Connection with Weighted Feature Fusion

被引:16
|
作者
Liu, Xiaoyu [1 ]
Li, Guo [1 ]
Chen, Wenkang [1 ]
Liu, Binghao [2 ]
Chen, Ming [1 ]
Lu, Shenglian [1 ]
机构
[1] Guangxi Normal Univ, Guangxi Key Lab Multisource Informat Min & Secur, Coll Comp Sci & Engn, Guilin 541004, Peoples R China
[2] Guangxi Acad Specialty Crops, Guangxi Citrus Breeding & Cultivat Engn Technol C, Guilin 541004, Peoples R China
来源
APPLIED SCIENCES-BASEL | 2022年 / 12卷 / 13期
基金
中国国家自然科学基金;
关键词
computer vision; citrus detection; YOLOv5; small objects; real-time detection; AGRICULTURE;
D O I
10.3390/app12136600
中图分类号
O6 [化学];
学科分类号
0703 ;
摘要
The accuracy detection of individual citrus fruits in a citrus orchard environments is one of the key steps in realizing precision agriculture applications such as yield estimation, fruit thinning, and mechanical harvesting. This study proposes an improved object detection YOLOv5 model to achieve accurate the identification and counting of citrus fruits in an orchard environment. First, the latest visual attention mechanism coordinated attention module (CA) was inserted into an improved backbone network to focus on fruit-dense regions to recognize small target fruits. Second, an efficient two-way cross-scale connection and weighted feature fusion BiFPN in the neck network were used to replace the PANet multiscale feature fusion network, giving effective feature corresponding weights to fully fuse the high-level and bottom-level features. Finally, the varifocal loss function was used to calculate the model loss for better model training results. The results of the experiments on four varieties of citrus trees showed that our improved model proposed to this study could effectively identify dense small citrus fruits. Specifically, the recognized AP (average precision) reached 98.4%, and the average recognition time was 0.019 s per image. Compared with the original YOLOv5 (including deferent variants of n, s, m, l, and x), the increase in the average accuracy precision of the improved YOLOv5 ranged from 7.5% to 0.8% while maintaining similar average inference time. Four different citrus varieties were also tested to evaluate the generalization performance of the improved model. The method can be further used as a part in a vision system to provide technical support for the real-time and accurate detection of multiple fruit targets during mechanical picking in citrus orchards.
引用
收藏
页数:17
相关论文
共 50 条
  • [1] Cross-scale feature fusion connection for a YOLO detector
    Ruan, Zhongling
    Wang, Hao
    Cao, Jianzhong
    Zhang, Hongbo
    IET COMPUTER VISION, 2022, 16 (02) : 99 - 110
  • [2] Learnable Cross-Scale Sparse Attention Guided Feature Fusion for UAV Object Detection
    Zuo, Xin
    Qi, Chenhui
    Chen, Yifei
    Shen, Jifeng
    Fan, Heng
    Yang, Wankou
    IEEE ACCESS, 2024, 12 : 114212 - 114226
  • [3] Cross-scale fusion detection with global attribute for dense captioning
    Zhao, Dexin
    Chang, Zhi
    Guo, Shutao
    Neurocomputing, 2021, 373 : 98 - 108
  • [4] Cross-scale fusion detection with global attribute for dense captioning
    Zhao, Dexin
    Chang, Zhi
    Guo, Shutao
    NEUROCOMPUTING, 2020, 373 : 98 - 108
  • [5] Image registration combining cross-scale point matching and multi-scale feature fusion
    Ou, Zhuolin
    Lu, Xiaoqi
    Gu, Yu
    CHINESE JOURNAL OF LIQUID CRYSTALS AND DISPLAYS, 2024, 39 (08) : 1090 - 1102
  • [6] Development of a cross-scale weighted feature fusion network for hot-rolled steel surface defect detection
    Zhang, Yuzhong
    Wang, Wenjing
    Li, Zhaoming
    Shu, Shuangbao
    Lang, Xianli
    Zhang, Tengda
    Dong, Jingtao
    ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE, 2023, 117
  • [7] YOLOv5 Transmission Line Fault Detection Based on Attention Mechanism and Cross-scale Feature Fusion
    Hao S.
    Yang L.
    Ma X.
    Ma R.
    Wen H.
    Zhongguo Dianji Gongcheng Xuebao/Proceedings of the Chinese Society of Electrical Engineering, 2023, 43 (06): : 2319 - 2330
  • [8] Anchor-free infrared pedestrian detection based on cross-scale feature fusion and hierarchical attention mechanism
    Hao, Shuai
    Gao, Shan
    Ma, Xu
    An, Beiyi
    He, Tian
    INFRARED PHYSICS & TECHNOLOGY, 2023, 131
  • [9] Cross-Scale Feature Fusion for Object Detection in Optical Remote Sensing Images
    Cheng, Gong
    Si, Yongjie
    Hong, Hailong
    Yao, Xiwen
    Guo, Lei
    IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, 2021, 18 (03) : 431 - 435
  • [10] An Improved YOLOv8-Based Lightweight Attention Mechanism for Cross-Scale Feature Fusion
    Liu, Shaodong
    Shao, Faming
    Chu, Weijun
    Dai, Juying
    Zhang, Heng
    REMOTE SENSING, 2025, 17 (06)