| 摘要: |
| 针对战机型号快速准确识别问题，提出一种利用梯度提升树的战机型号快速识别方法。以多传感器融合的战机航迹解译信息为数据基础，通过分析航迹数据特征，构建战机航迹数据特征工程，利用boosting集成学习思想，训练基于梯度提升决策树的战机型号分类器，可准确识别每个航迹点对应的战机型号。实测数据实验结果表明，所提模型识别准确率达到95.76〖WT《Times New Roman》〗%〖WTBZ〗，较卷积神经网络（Convolutional Neural Network,CNN）方法识别准确率90.32〖WT《Times New Roman》〗%〖WTBZ〗提高了5.44〖WT《Times New Roman》〗%〖WTBZ〗；所提模型平均单点识别计算耗时为408.1 μs，较CNN方法耗时5 385.5 μs快13.19倍，证明了所提算法能够快速有效地辨识战机型号，满足准确性和实时性需求。 |
| 关键词: 目标识别 战机型号识别 机器学习 梯度提升树 |
| DOI： |
|
| 基金项目: |
|
| Fast recognition of aircraft types with gradient boosting decision tree |
| ZHAO Wenbin |
| (Southwest China Institute of Electronic Technology,Chengdu 610036,China) |
| Abstract: |
| To fast recognize aircraft types，a method based on gradient boosting decision tree is proposed.Specifically,the aircrafts’ traces,interpreted by the results from multisensors fusion,are used as the raw data source.Then,after data are cleaned and labelled,multiple elementary features are designed by analyzing the data characteristics for seeking an appropriate expression of statistic distribution of the data.Finally,a classifier based on gradient boosting decision tree is designed,which has the capability to recognize aircraft types at each trace point time instance.The realdata experimental results illustrate that the trained model using the proposed method achieves the accuracy of 95.76〖WT《Times New Roman》〗%〖WTBZ〗,which is 5.44〖WT《Times New Roman》〗%〖WTBZ〗 higher than that of the convolutional neural network(CNN) model,and the average calculation time cost of a single trace point is tested as 408.1 μs,which is 13.19 times faster than that of the CNN model,5385.5μs,proving the good performance of the proposed method both on the classification accuracy and the real time. |
| Key words: target recognition aircraft type recognition machine learning gradient boosting decision tree |
