论文部分内容阅读
一维距离像是宽带雷达目标识别的重要特征之一。本文根据弹道目标的微动特性,推导了微动弹道目标的时间-距离像模型。然后提取具有平移不变性的中心矩和双谱作为待识别特征向量,并分别使用K-L变换和局部双谱法对提取到的中心矩和双谱特征进行降维。将降维后的特征分别输入支持向量机进行分类识别,最后将支持向量机的输出进行决策级融合,得到待识别目标的识别概率。与基于单特征量的识别方法相比,本文提出的方法不仅具有较高的识别率,而且具有良好的抗噪能力。
One-dimensional range image is one of the important features of wideband radar target recognition. In this paper, based on the fretting characteristics of the trajectory target, the time-distance model of the fretting trajectory target is deduced. Then, the center moments and bispectrum with translational invariance are extracted as the eigenvectors to be identified, and the extracted central moments and bispectrum features are dimensionally reduced by K-L transform and partial bispectrum respectively. The dimensionality-reduced features are input into SVM for classification and recognition respectively. Finally, the output of SVM is merged into decision-making level to get the recognition probability of the target to be recognized. Compared with the identification method based on single feature, the proposed method not only has high recognition rate, but also has good anti-noise ability.