Urinalysis plays a vital role in diagnosing kidney diseases and urinary tract infections, but conventional manual analysis is often subjective, time-intensive, and prone to variability. Existing detection and segmentation methods struggle with accurately identifying small and irregularly shaped urine sediment particles, leading to inconsistencies in results. To overcome these limitations, we introduce YOLOv11-SAMNet, a novel one-stage instance segmentation model that combines the YOLOv11 detector with the Segment Anything Model (SAM) for 14 classes. Our approach enhances both detection precision and segmentation accuracy by leveraging a feature extraction backbone for capturing key image details, a multi-scale fusion neck for integrating diverse feature representations, and a detection and segmentation head to refine instance segmentation, followed by a bounding box. The model performs exceptionally well on well-defined structures, such as Leukocytes (94.3% mAP@50) and Epithelial cells (92.5% mAP@50), but encounters challenges when segmenting small and irregular elements like bacteria (33.8% mAP@50) and mucus (27.2% mAP@50). To further enhance model performance, we developed an expert-assisted data preprocessing pipeline to improve the quality of training data. Our findings highlight the potential of YOLOv11-SAMNet in automating urinalysis, offering a more efficient and objective alternative to traditional manual analysis, ultimately contributing to improved clinical diagnostics.