| 摘要: |
| 针对战场空中飞行目标型号辨识不清的问题，提出了一种采用长短时记忆（Long Short-Term Memory，LSTM）循环神经网络的战场空中飞行目标型号辨识方法。以飞行目标航迹态势数据为数据源，分析数据不同特征特点，设计增加基于数据原有特征的派生特征，构建并训练基于LSTM循环神经网络的目标型号分类器，形成目标型号序列化智能辨识模型，实现仅依赖于航迹的战场空中飞行目标型号时序化准确辨识。经实测数据验证，所提方法识别准确率达到96.73%，F1分数达到0.968 3，较梯度提升树模型方法识别准确率至少提高了1.02%，而F1分数至少提高了0.013 2，证明了所提方法的有效性。 |
| 关键词: 战场空中飞行目标 目标识别 型号辨识 机器学习 循环神经网络 长短时记忆网络 |
| DOI： |
|
| 基金项目: |
|
| Battlefield aircraft types recognition with recurrent neural network |
| WANG Mingyang |
| (Southwest China Institute of Electronic Technology, Chengdu 610036, China) |
| Abstract: |
| The recognition problem for battlefield aircraft types is addressed.The categories are recognized accurately with the proposed long short-term memory(LSTM) recurrent neural network(RNN).Specifically,the aircrafts’ traces,which are situation data,are chosen as the raw data source.Then,after data cleaning and labelling,the derived features are designed by analyzing the raw data features to enhance the features.Finally,a classifier based on LSTM RNN is constructed and trained,which has capability to accurately recognize the categories of aircrafts in a serialization manner only with the traces.The real-data experimental results illustrate that the proposed model achieves the accuracy of 96.73% and the F1 score of 0.968 3,which is at least 1.02% and 0.013 2 higher than those of the method with gradient boosting tree respectively,proving the good performance of the method. |
| Key words: battlefield aircraft target recognition type identification machine learning recurrent neural network long short-term memory |
