LVQ神经网络用于海上小目标检测

Application of LVQ neural network to small sea target detection

舰载(或岸基)雷达对海面小目标检测是一个重要而困难的课题。最大幅度极点检测(ARLPM)是目前实用性较强的一种方法,特别是在短时检测方面有着不错的效果。本文在AR-LPM的基础上提取了最大幅度极点的两个目标特征———频率和谱宽,然后采用学习矢量量化(LVQ)神经网络进行目标检测。该方法有强大的非线性运算和相似特征聚类功能,不仅简单易行,还具备一定的自适应性。最后本文利用真实海杂波数据对该方法作了实验,通过比较表明,LVQ神经网络检测优于原方法。

Small targets detection in sea clutter is an important and difficult problem for shipborne （or shore-based） radars. The autoregressive largest pole magnitude （ARLPM） is a method with high practicability, especially for short time detection. Yet the shortcoming of it is that only the pole＇s spectral width of the AR spectral is used. Two characters as frequency and spectral width of autoregressive largest pole based on ARLPM are obtained. The learning vector quantization （LVQ） neural network is used to detect targets. Being very capable of making nonlinear operation and similar character clustering, this method is easy to operate and self-adaptive. A simulation is made for this method with the data of true sea clutter. The comparison shows the method using LVQ neural network is better than the ARLPM.