The flight parameters of the warhead fragment group are of great significance for the assessment of destructive power. Aiming at the problems of low detection accuracy and insufficient feature expression of fragmented small targets in complex backgrounds

a lightweight detection model based on multi-scale feature fusion and hybrid attention mechanism YOLO (MSMA-YOLO) is proposed. The multi-scale spatial pyramid pooling fast (MS-SPPF) module is introduced on the YOLOv8 framework to enhance the feature aggregation ability through the multi-scale attention mechanism. Design a lightweight mixed attention two convolutions fusion (MA-C2f) module to optimize the fusion of shallow and deep features; And combined with dynamic convolution (DyConv) to achieve dynamic offset learning

in order to

improve the integrity and adaptability of feature representation. The experimental results on the self-made fragment dataset show that the precision rate

recall rate and mAP

0.5

of the MSM-YOLO model reach 76.2%

71.2% and 72.2% respectively

which are 4.3%

2.5% and 2.8% higher than those of the baseline model YOLOv8n. The model parameters only increase by 0.363×106 and the volume increases by approximately 0.8 MB. The proposed model achieves an excellent balance between detection accuracy and computational efficiency

providing a reliable target detection basis for the extraction of flight parameters of fragments and damage assessment.