Improvement of YOLOv7 with Attention Modules for Urinary Sediment Particle Detection

Urinary sediment analysis is crucial for assessing kidney health. Traditional machine learning techniques treat urine sediment particle detection as an image classification problem, but it is challenging due to low contrast and weak edges. To improve detection and focus on particles, attention-based urinary sediment detectors have been developed. This paper proposes a YOLOv7-based urinary sediment detection with attention modules, enhancing feature extraction and reducing background noise by locating attention modules in backbone part. Experimental results obtained from a urinary sediment dataset demonstrate that our proposed models outperform both the original YOLOv7 and YOLOv5s-CBL, which is a state-of-the-art urinary sediment detector, in terms of recall and mAP. Our models achieve significantly higher recall scores, surpassing the original YOLOv7 by 12.4% and the compared detector by 4.3%. Moreover, our models exhibit improved mAP scores, surpassing the original YOLOv7 by 7.4% and the compared detector by 1.6%.