基于高分辨距离像的雷达目标识别是该领域的研究热点，而特征提取是其中的关键环节。相对于散射中心强度和位置特征而言，长度特征随目标方位角变化的影响较小，是一种相对稳定的特征，而且长度特征属于目标本身的物理特征，具有实际的物理意义。文中提出一种基于双向滑动平均的目标长度特征提取方法，并将提取的长度特征用于目标粗分类。该方法首先对距离像进行降噪处理，然后从左右两端同时向中间进行滑动平均处理，当滑动均值大于预设的阈值时，即可确定目标区域的起始和终止位置，从而得到目标的长度特征。该方法的核心在于分别估计左右两端各自的阀值，而不是采用一个统一的阀值，并且在阀值估计的过程中同时考虑了距离像均值和噪声的影响。因此，该方法对于距离像突变、非目标区域野值等具有较强的稳健性。通过对五类飞机的实测数据进行实验，验证了该方法的有效性。

Radar automatic target recognition based on high range resolution profile (HRRP) has been subject to intense research and feature extraction is a critically important process. Generally, a number of dominant scatterers are selected from a HRRP and their intensities and locations are used as a target feature. However, both intensity and location are in fast variation with the changes of the target aspect angle and this degrades the classification performance. By contrast, the target length, as an inherent physical feature of target, varies relatively slowly with the aspect angle, thus providing a promising feature for rough classification purpose. In this paper, a two way sliding average method is proposed to extract the target length feature from a HRRP. The HRRP is firstly pre-processed by using a de-noising method, and then the sliding average method is conducted on the pre-processed HRRP from the left and right side simultaneously. The beginning or ending position of target area is determined and the target length is obtained when the sliding mean goes over the preset threshold. The core of our method lies in that two individual thresholds are estimated for the left and right side respectively rather than one single. Moreover, by taking both the mean of the HRRP and noises into consideration, the method is relatively robust to abrupt changes in the HRRP or outliers in non-target area. Experimental results based on measured data from five airplanes show the effectiveness of the proposed method.

TN911.7