Small Target Detection in UAV Images Based on NDM-YOLOv8
In view of the phenomenon that there are many instances of small targets in UAV aerial images and there are occlusions between targets,which can easily cause problems such as missed detection and false detection,a new small target detection algorithm based on a non-strided dynamic multihead structure,NDM-YOLOv8(Non-strided Dynamic Multihead YOLOv8),is proposed.Firstly,the SConv module was designed,which integrated non-strided convolution and preserved the discriminative feature information of the input data as much as possible to reduce the loss of fine-grained information of small target features.Secondly,the C2f-LSK module was designed to effectively weight spatial features through a selection mechanism,dynamically adjust receptive fields,flexibly capture features and contextual information of different scales,and improve the model's attention to small targets.Finally,a P2 small target detection head was designed and residual connected to the high-level network to reduce the loss of small object features and enhance the algorithm's ability to extract small target features.Experiments show that the NDM-YOLOv8 effectively improves the detection accuracy of small targets in drone images.On the public dataset VisDrone2019,the NDM-YOLOv8 improved by5.3 percentage points on mAP0.5 and 3.3 percentage points on mAP0.5:0.95 compared with YOLOv8n.Compared with other models,it also achieved better detection performance and can more effectively complete the task of small target detection in UAV aerial images.
国家科技期刊平台
NSTL
万方数据
