Abstract: 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．