南京理工大学学报（自然科学版）2026，Vol.50Issue(2)：183-194,12.DOI:10.14177/j.cnki.32-1397n.2026.50.02.008
基于YOLOv7-SEFNet的碎片群目标检测算法研究
Research on fragment cluster target detection algorithm based on YOLOv7-SEFNet
摘要
Abstract
In the context of high-speed photography as a non-contact method for fragment cluster-behind-target parameter testing,fragment cluster target detection serves as a critical step in parameter testing.To address the low contrast and unclear targets in fragment cluster images,this paper proposes a multi-scale fusion image enhancement algorithm combining Gamma correction and an unsharp mask to improve brightness and detail of fragment images.To address issues such as missed and false detections in small target detection,this paper introduces the fine-sensitive YOLOv7 algorithm(YOLOv7-SEFNet),which replaces CIoU loss function with SIoU for better bounding box regression accuracy,constructs a small target detection layer to improve the network's adaptability to minute features,and integrates an attention mechanism module to enhance the weighting focus on small targets.Experiments show that YOLOv7-SEFNet improves the recall rate by 3.86%,the precision rate by 3.09%,and the average precision rate by 1.77%over the original YOLOv7.When compared to networks ranging from YOLOv4 to YOLOv13,YOLOv7-SEFNet demonstrates superior overall performance in the task of fragment target detection,effectively enhancing the detection accuracy for images of fragment cluster-behind-target.关键词
靶后碎片群/多尺度融合/图像增强/小目标检测/YOLOv7-SEFNetKey words
fragment cluster-behind-target/multi-scale fusion/image enhancement/small target detection/YOLOv7-SEFNet分类
信息技术与安全科学引用本文复制引用
付欣荣,孔筱芳,徐春冬,罗红娥,弯港,夏言,万敏杰..基于YOLOv7-SEFNet的碎片群目标检测算法研究[J].南京理工大学学报（自然科学版）,2026,50(2):183-194,12.基金项目
国家自然科学基金青年科学基金项目(62201260) （62201260）
国家自然科学基金面上项目(62571245) （62571245）
中央高校基本科研业务费专项资金(30923011015 （）
30925020226 （）
30924010941) （）
江苏省自主科研基金项目(2025-JSS-LB-034-14) （2025-JSS-LB-034-14）
装备预研武器工业应用创新项目(627010402) （627010402）
