基于LSTM-Attention的空间目标分类研究
Research on space object classificationbased on LSTM-Attention
杨礼友 1余显冰 1李智1
作者信息
- 1. 四川大学电子信息学院,成都 610065
- 折叠
摘要
针对空间目标特有属性及其运动趋势难以使用单一元素进行描述,以及现有空间目标分类技术准确度低等问题,构建了一种基于LSTM-Attention的空间目标分类模型,该模型无需开展额外的特征工程,能够联系空间目标序列数据的上下文信息和长期依赖关系,提取样本的局部特征并对其长期运动趋势进行建模.利用Mini-Mega TORTORA(MMT)系统实测光变曲线进行验证,与传统方法相比,设计使用的模型拥有较高的数据处理效率,能够提高空间目标的分类准确度并满足空间态势感知的部分应用需求.
Abstract
Aiming at the problems of the difficulty of describing the unique attributes of space objects and their motion trends using a single element,as well as the low accuracy of the existing space object classification techniques,constructs a space object classification model based on LSTM-Attention.The model does not need to carry out additional feature engineering,and is capable of linking the contextual information and long-term dependencies of space object sequence data,extracting the local features of the object and modelling its long-term motion trends.Validated using real light curve observations taken from the Mini-Mega TORTORA(MMT)system,designed to use a model with high data processing efficiency compared to traditional methods,and is able to improve the classification accuracy of space objects and satisfy some of the application requirements of space situational awareness.
关键词
空间目标分类/深度学习/LSTM/注意力机制Key words
space object classification/deep learning/LSTM/attention mechanism引用本文复制引用
出版年
2024
NETL
NSTL
万方数据