基于大场景合成孔径雷达（SAR）图像序列，研究了一种针对多类慢动车辆目标的识别与跟踪方法，采用先识别、再跟踪的思路。提出了一种图像目标局部多分辨分析与多核分类器相结合的识别方法，实现了多类目标的快速特征提取和准确分类。根据相邻帧之间目标的对应关系，利用无偏卡尔曼滤波对目标的运动参数进行估计，并用实际测量值不断进行修正，实时获取目标的坐标、类型等信息，实现了复杂背景下地面多类慢动目标的高效跟踪。通过构建大场景合成孔径雷达序列图像进行仿真实验，证实了该方法具有快速和稳定的收敛性能，实时性较好，具有较高的跟踪精度。

Based on the large-scale scene synthetic aperture radar (SAR) image series, a combined recognition and tracking method for various slowly-moving vehicle targets is presented. The method introduces an idea named recognition first and tracking later. A recognition algorithm combining local multi-resolution analysis for image target and multiple kernel classifier is studied, which realizes the high-speed feature extraction and accurate classification of various image targets. According to the corresponding relationship of targets in the adjacent frames, the targets′ motion parameters are estimated utilizing the unscented Kalman filter. Simultaneously, the real-time location and target type can be obtained via continuous correction using the measured values; as a result, high performance tracking of various slowly-moving ground targets in complicated background is realized. Large-scale scene SAR series images are constructed and several simulation tests are performed, demonstrating that the method has good convergence, excellent real-time performance, and high tracking precision.