基于概率神经网络的通用航空器冲突探测方法

A Method to Detect General Aircraft Collisions Based on Probabilistic Neural Network

针对当前冲突探测技术难以同时实现精准识别与实时识别的问题,研究基于概率神经网络(PNN)的通用航空器冲突探测方法。将冲突探测视为模式识别问题,通过冲突模型分析,提出了航空器“冲突角”概念,改进了现有冲突识别方法采用的关键特征指标,将原有的4个关键特征指标提炼为3个指标,分别为航空器相对距离、相对速度以及冲突角,以此构造概率神经网络,训练形成神经网络分类器。结果表明,基于3关键特征的概率PNN冲突分类器分类误警率和漏警率保持在1%左右,在冲突误警率上优于基于4特征的SVM冲突分类器的6%,提高了航空器冲突探测的准确度;分类所耗时间始终保持在1.2s左右,远低于Monte Carlo仿真方法的同时,较4特征分类器也降低了0.2s左右,提高了冲突识别效率。

A Probabilistic Neural Network(PNN)based conflict detection technology for general aircrafts is studied to achieve accurate and real-time recognition at the same time.Problems of conflict detection are regarded as a pattern recognition problem.A concept of“conflict angle”is introduced into characteristic indicators adopted by existing conflict recognition methods.Traditional four characteristic indicators are extracted into three indicators,which are relative distance,relative speed,and conflict angle.Then a PNN network is constructed and a classifier is trained.The results show that false alarm rate and missed alarm rate of the PNN conflict classifier based on three indicators are about 1%,which are better than SVM conflict classifier based on four indicators.And the time for classification is about 1.2 s,which is much lower than Monte Carlo simulation.Compared with the four-characteristic classifier,the time is also reduced by about 0.2 s.The accuracy and efficiency of conflict detection can be improved.