Abstract: Under complex garden environment such as aggregation and occlusion， it is difficult for the existing target detection algorithms to accurately detect spherical hedge. In order to solve this problem， an algorithm YOLO-CBS based on YOLOv5s is proposed to improve the detection accuracy of spherical hedge in gardens. Firstly， coordinate attention （CA） is introduced into the backbone network of YOLOv5s， which considers the relationship between channels and the location information of the feature space， so that the model can more accurately identify and locate the target hedge. Secondly， the path aggregation network （PANet） is replaced by bidirectional feature pyramid network （BiFPN） to improve the efficiency of feature fusion. Finally， the non‑maximum suppression （NMS） at the output is changed to Soft-NMS to improve the detection accuracy of target hedges under complex scenes such as occluded hedges and dense hedges. The results of experiments on a typical hedgerow data set show that the average accuracy of the YOLO-CBS algorithm is improved by 3.4% compared to the YOLOv5s algorithm.

Key words:  , hedge detection； YOLOv5s； attention mechanism； feature pyramid network； non?maximum suppression

摘要： 在聚集遮挡等复杂园林环境下，现有的目标检测算法很难对球形绿篱进行准确检测。针对这一问题，提出一种基于YOLOv5s的算法YOLO-CBS，用于提高园林球形绿篱的检测精度。首先，将坐标注意力（CA）引入YOLOv5s的主干网络，CA不仅考虑通道间的关系还考虑特征空间的位置信息，因而能够使模型更准确地识别和定位目标绿篱；其次，用双向特征金字塔网络（BiFPN）替换路径聚合网络（PANet），以提高特征融合的效率；最后，将输出端的非极大值抑制（NMS）改为Soft-NMS，以提高对遮挡绿篱、聚集绿篱等复杂场景下的目标绿篱检测精度。典型绿篱数据集试验结果表明，与YOLOv5s算法相比，YOLO-CBS算法平均精度提高3.4%。

关键词: 绿篱检测, YOLOv5s, 注意力机制, 特征金字塔网络, 非极大值抑制