基于几何散列法的ISAR像自动目标识别

摘要

Inverse synthetic aperture radar (ISAR) images, different from the optical images and consisting of many sparse scatterers, usually vary with different view angles and suffer jamming and shelter, which make ISAR targets classification difficult. An available recognition method based on geometric hash is presented. The approach extracts key points which reflect the structure of ISAR targets at first, then constructs affine coordinates and obtains the affine-in variance features to solve the variation problem of translation, rotation and scale. Finally, geometric hash is applied as the final classifier, which performs excellently in anti-jamming and local recognition, to solve the problem of location and intensity variation of scatterers due to pose alternation, jamming and shelter. Experimental results show that the proposed method is quite effective in distinguishing targets with different structures and performs well in local recognition.%逆合成孔径雷达(inverse synthetic aperture radar,ISAR)像不同于一般光学像,通常表现为随视角变化的稀疏散射中心分布,且存在干扰或遮挡现象,这使ISAR目标识别存在很多困难.针对上述问题,提出一种基于几何散列法的ISAR像识别方法.首先获取能反映目标结构信息的特征点；然后利用特征点之间的几何关系构造仿射坐标,获取目标的仿射不变量,以解决目标成像视角变化引起的图像平移、旋转和尺度变化等问题；最后针对姿态敏感性、干扰或遮挡导致的散射点位置和强度的变化问题,采用具有良好的抗干扰和局部识别性能的几何散列法来完成识别.仿真实验表明,该方法能够有效区分不同结构的目标,且对干扰或遮挡现象具有良好的局部识别性能.