Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (4): 237-246.DOI: 10.3778/j.issn.1002-8331.2107-0523

• Graphics and Image Processing • Previous Articles     Next Articles

Improved YOLOv5 Method for Detecting Growth Status of Apple Flowers

YANG Qisheng, LI Wenkuan, YANG Xiaofeng, YUE Linxi, LI Haifang   

  1. College of Information and Computer, Taiyuan University of Technology, Jinzhong, Shanxi 030600, China
  • Online:2022-02-15 Published:2022-02-15



  1. 太原理工大学 信息与计算机学院,山西 晋中 030600

Abstract: Aiming at the problem that it is difficult for the existing target detection algorithm to detect the growth status of apple flowers in the complex environment of the orchard, an improved YOLOv5 method for detecting the growth status of apple flowers is proposed. The growth status of the flowers during the flowering period of the tree is tested. This method firstly improves the cross-stage local network module, adjusts the number of modules, and designs the backbone network with the collaborative attention module to improve model detection performance and reduce parameters. Secondly, this method combines the new detection scale and the split-based convolution operation to design a feature fusion network to improve the network feature fusion ability, and finally chooses CIoU as the loss function of the frame regression to achieve high-precision positioning. The improved algorithm is compared with the original YOLOv5 algorithm on a self-built data set. The experimental results show that the mAP of the algorithm in this paper reaches 0.922, which is 5.4 percentage points higher than that of YOLOv5. Compared with other mainstream algorithms, the detection accuracy is greatly improved, which proves the effectiveness of the algorithm.

Key words: YOLOv5, agricultural automatic monitoring, feature fusion, object detection

摘要: 针对现有目标检测算法难以在果园复杂环境下对苹果花朵生长状态进行高精度检测的问题,提出一种改进YOLOv5的苹果花朵生长状态检测方法,对花蕾、半开、全开、凋落四类苹果树开花期花朵生长状态进行检测。该方法对跨阶段局部网络模块进行改进,并调整模块数量,结合协同注意力模块设计主干网络,提高模型检测性能并减少参数。结合新的检测尺度与基于拆分的卷积运算设计特征融合网络,提升网络特征融合能力。选用CIoU作为边框回归的损失函数实现高精度的定位。将改进算法与原始YOLOv5算法在自建数据集上进行对比实验,结果表明,改进算法mAP达到0.922,比YOLOv5提高5.4个百分点,与其他主流算法相比检测精度有较大提升,证明了算法的有效性。

关键词: YOLOv5, 农业自动监测, 特征融合, 目标检测