相似目标识别的投影子空间构造方法

基于值空间投影的合成孔径雷达目标识别技术要求不同目标的值空间的交集应尽可能小。然而，对具有相似形态及结构的目标进行识别时，相应目标的值空间存在较大的交集部分，影响识别性能。通过约束投影空间与值空间的关系，提出一种有效构造投影空间的方法。该方法可以增大相似目标在投影空间投影的差异，提高目标识别性能。实验结果表明，本文方法构造的投影空间增大了不同目标间的可分性，识别性能及识别效率明显优于其他经典识别方法。

Method of projection subspace construction for recognizing similar targets

Automatic target recognition technology, which is based on the projection on the range subspace of synthetic aperture radar images, requires that the intersection between the range subspaces of the different targets should be minimal. However, to distinguish those targets with similar appearance and structure, the intersection between the corresponding range subspaces is big which results in performance of recognition degradation. By using the constraint between the projection subspace and the range subspace, an efficient method for constructing the projection subspace is proposed. The proposed method can enlarge the difference between the projections of the similar targets on the projection subspace and improve the performance of target recognition. Experimental results show that the projection subspace constructed by the proposed method increases the separablity between the different similar targets, and the recognition performance and efficiency are superior to other classical identification methods.