English

中国农机化学报

中国农机化学报 ›› 2024, Vol. 45 ›› Issue (9): 190-195.DOI: 10.13733/j.jcam.issn.2095-5553.2024.09.029

• 农业信息化工程 • 上一篇    下一篇

基于改进 YOLOv4的蔗种坏芽识别方法研究

沈漫林,刘姣娣,许洪振,何捷,段玉龙   

  1. (桂林理工大学机械与控制工程学院,广西桂林,541006)
  • 出版日期:2024-09-15 发布日期:2024-09-04
  • 基金资助:
    国家自然基金项目(52265028);广西自然科学基金项目(2021JJA160046);广西研究生教育创新计划项目(YCSW2022333)

Research on sugarcane seed bad bud recognition method based on improved YOLOv4

Shen Manlin,Liu Jiaodi,Xu Hongzhen,He Jie,Duan Yulong   

  1. (School of Mechanical and Control Engineering,Guilin University of Technology,Guilin,541006,China) 
  • Online:2024-09-15 Published:2024-09-04

摘要:

为实现蔗种切种机构对坏芽蔗种实时检测剔除,提出一种基于改进 YOLOv4的蔗种坏芽快速识别方法。通过在 YOLOv4主干网络添加轻量的注意力模块(CBAM),以增强网络提取蔗芽特征能力,降低背景噪声对蔗芽识别精度的影响;并利用 K-means算法对数据集重新聚类,生成符合蔗芽特征的锚定框,提高蔗种坏芽检测精度;将路径聚合网络中原有的标准卷积替换为深度可分离卷积,大幅减少参数降低计算负荷,整体识别速度得到提升。测试结果表明:改进后的网络模型比 YOLOv4精确率提高 3. 12%,平均精确率均值提高 4. 15%,召回率提高 3. 69%,单张图像识别时间缩短 7 ms。改进后算法实现对蔗种坏芽的快速准确识别,满足切种机构实时检测并剔除蔗种坏芽的需求。

关键词: 蔗种坏芽, 改进网络, 切种机构, 注意力模块, 聚类算法, 深度可分离卷积 ,

Abstract:

In order to realize the real.time detection and elimination of the bad buds by the sugarcane seed cutting mechanism,a rapid recognition method for the bad buds of sugarcane seed based on improved YOLOv4 was proposed. A lightweight Convolutional Block Attention Module(CBAM)was added to the YOLOv4 backbone network to enhance the ability of sugarcane bud feature extraction and reduce the influence of background noise on the accuracy of sugarcane bud recognition. The K-means algorithm was used to re.cluster the data set to generate an anchor frame that was consistent with the characteristics of cane buds,which improved the detection accuracy of cane seed bad buds. The original standard convolution in the path aggregation network is replaced by deep separable convolution, which greatly reduces the parameters and computational load,and improves the overall recognition speed. The training and test results show that compared with YOLOv4, the precision of the improved network model is increased by 3. 12%, the mean average precision is increased by 4. 15%,the recall is increased by 3. 69%,and the recognition time of single image is shortened by 7 ms. The improved algorithm realized rapid and accurate identification of sugarcane seed bad buds and met the need of real.time detection and removal of sugarcane seed bad buds by the seed cutting mechanism. 

Key words: sugarcane seed bad bud, improved the network, seed cutting mechanism, attention module, clustering algorithm, depth separable convolution ,

中图分类号: