航空兵器2025，Vol.32Issue(5)：104-113,10.DOI:10.12132/ISSN.1673-5048.2025.0064
基于扩散模型和时序数据图像化的飞机机动识别方法
Aircraft Maneuver Recognition Method Based on Diffusion Model and Time Series Data
彭天昊 1吴达 1张杨子1
作者信息
- 1. 空军工程大学防空反导学院,西安 710051
- 折叠
摘要
Abstract
Aircraft maneuver recognition plays a critical role in quantifying pilot training effectiveness,predicting tactical intentions of the opponents,and securing battlefield dominance.However,the severe imbalance in battlefield data substantially hinders the practical deployment of this technology.Recent advancements in generative artificial in-telligence,particularly the denoising diffusion probability model(DDPM),have demonstrated exceptional sample ge-neration capabilities in computer vision applications.Drawing inspiration from this progress,this paper proposes a Mar-kov transition field(MTF)-based time-series visualization framework.By encoding aircraft maneuver time-series data into 2D grayscale images and leveraging DDPM for synthetic sample generation,this approach effectively mitigates data imbalance while reformulating time-series classification as an image classification task.Therefore,this paper designs a novel hybrid classification architecture that synergistically integrates MobileNetV3's efficient local feature extraction ca-pability with Swin-Transformer's global contextual attention mechanisms.It constructs an aircraft maneuver recognition method that combines visualization methods,DDPM generation models and classification networks.Experimental vali-dation demonstrates that this method achieves significantly higher accuracy in aircraft maneuver identification tasks compared to other classical models in the field of image classification.关键词
机动识别/去噪扩散概率模型/样本不平衡/马尔可夫转移场/MobileNetV3/Swin-TransformerKey words
maneuver recognition/denoising diffusion probability model/sample imbalance/Markov transfer field/MobileNetV3/Swin-Transformer分类
武器工业引用本文复制引用
彭天昊,吴达,张杨子..基于扩散模型和时序数据图像化的飞机机动识别方法[J].航空兵器,2025,32(5):104-113,10.
