基于多维特征的舰载机舰面作业识别

Carrier aircraft deck operation recognition based on multi-dimensional feature

  • 摘要: 【目的】航母舰面空间有限、布局复杂,舰载机舰面作业受环境因素约束,且不同作业参与个体数量和作业时间不同,个体间时空交互密切程度动态改变。为此,提出一种基于多维特征的舰载机舰面作业识别方法。【方法】首先,精准选取航道边界和静态障碍物等关键点表征环境信息,并通过图卷积网络建模动态个体与静态环境对象的交互关系,深度挖掘作业对象交互的潜在联系。然后,设计多尺度时空特征提取模块,引入扩张注意力机制,通过设置不同的扩张率关注全局和局部空间中关键个体交互关系;同时,采用时序卷积网络(TCN)和注意力机制提取时间维度上个体间的交互特征,从而有效捕捉个体间长短序的动态关系;最后,堆叠多个设计的多尺度时空特征提取模块,自适应提取多维度特征,提高舰载机舰面作业识别准确率。【结果】实验结果表明,在自建的不同视角异构个体舰载机舰面作业识别数据集上,所提方法的准确率相较于ARG,DIN,AT,GroupFormer等群体活动识别方法有明显提升,实现了97.83%的识别精度。【结论】该方法能够实现舰载机舰面作业的高精度识别,可为提升航母作战效能提供重要支撑。

     

    Abstract: Objectives Carrier aircraft deck space is limited, the layout is complex, the carrier aircraft deck operation is subject to environmental factors constraints, and the number of individuals involved in different operations and operating time is different, and the degree of close spatial and temporal interaction between individuals dynamically changes. To this end, a multi-dimensional feature-based recognition method for carrier aircraft deck operation is proposed. Methods Firstly, key points such as channel boundaries and static obstacles are accurately selected to represent the environmental information, and the interactions between dynamic individuals and static environmental objects are modelled by graph convolutional networks, to deeply explore the potential connections of operational object interactions. Then, a multi-scale spatio-temporal feature extraction module is designed to introduce the dilated attention mechanism, which focuses on the key individual interactions in global and local space by setting different dilation rates; at the same time, temporal sequential convolutional networks (TCN) and the attention mechanism are used to extract the interaction features between individuals in the temporal dimension, so as to efficiently capture the dynamic relationships between individuals in the long and short sequences; finally, multiple designed multi-scale spatio-temporal feature extraction modules are stacked to adaptively extract multi-dimensional feature to improve the recognition accuracy of carrier aircraft deck operations. Results Experimental results show that, on a self-built dataset of heterogeneous object carrier aircraft deck operation recognition from different perspectives, the proposed method significantly outperforms group activity recognition methods such as ARG, DIN, AT, and GroupFormer in terms of accuracy, achieving a recognition precision of 97.83%.Conclusions The proposed method achieves high precision in carrier aircraft deck operation recognition and provides important support for enhancing aircraft carrier operational effectiveness.

     

/

返回文章
返回