基于改进YOLOv5模型的茄科植物叶片实例分割方法

Instances Segmentation Method for Solanaceae Plant Leaf Based on Improved YOLOv5 Model

观测叶片是了解植物生长情况的重要措施,为实现温室系统智能化管理,确保茄科植物健康生长,使用实例分割技术可以获取到茄科植物在植物苗期的叶片生长信息。提出一种基于YOLOv5模型的茄科植物叶片实例分割模型YOLOv5-Biformer,该模型针对茄科植物叶片的小目标特征,在主干网络中加入稀疏注意力网络,可以有效提高茄科植物叶片实例分割效率。试验结果表明,YOLOv5-Biformer模型在茄科植物叶片数据集上与基准模型相比,在精确度、召回率和平均精度指标上分别提高0.5、1.9和1.0个百分点。该模型在智能温室环境下对于苗期茄科植物叶片的实例分割有显著效果,为实现温室系统智能化管理提供新思路。

Observing leaves is an important measure to understand plant growth.To achieve intelligent management of greenhouse systems and ensure healthy growth of solanaceous plants,instance segmentation techniques can be used to obtain leaf growth information of solanaceous plants during seedling stage.A solanaceous plant leaf instance segmentation model called YOLOv5-Biformer was proposed based on YOLOv5 architecture.This algorithm addressed characteristics of small targets,i.e.,solanaceous plant leaves,and incorporated a sparse attention network into backbone network to effectively improve efficiency of solanaceous plant leaf instance segmentation.Experimental results indicated that,YOLOv5-Biformer model improved accuracy,recall and average accuracy indicators by 0.5,1.9 and 1.0 percentage points respectively compared to benchmark model on solanaceae plant leaf dataset.This network model demonstrated significant effectiveness in instance segmentation of solanaceous plant leaves during seedling stage in context of an intelligent greenhouse environment.This research could provide new insights for achieving intelligent management of greenhouse systems.