Abstract: Hu moment invariants，describe the characteristics of the region，are widely used in pattern recognition.However，due to image noise，especially speckle noise in SAR images，Hu moment invariants no longer maintain its perfect performance.Accordingly，in this paper，a novel feature vector which combines Hu moment invariants and four affine moment invariants and gray peak，mean and variance coefficient of target area in SAR images is proposed.The feature vector not only describes target regional shape but also characterizes gray statistical distribution of local target region in SAR images.The simulations of target recognition and classification in two different resolution SAR images preferably indicate that the feature vector is effective and acquires high target recognition performance.

Key words: SAR image, feature extraction, classification and recognition

摘要： 用于描述区域特征的Hu矩不变量在模式识别中得到广泛使用。然而在噪声影响下，尤其是SAR图像中严重的相干斑噪声，Hu 矩不变量不再保持其完美的性能。以Hu七个矩不变量为基础，结合SAR图像的特点，引入四个仿射矩不变量和SAR图像中目标区域的峰值、均值和方差系数，构成SAR图像中目标识别的特征向量。该特征向量体现了SAR图像区域目标的形状特征和区域的灰度信息。通过对两种不同分辨率下的T72坦克SAR图像的目标识别仿真实验，均获得了较好的目标识别效果，说明所选取的SAR图像目标识别的特征向量是有效的，具有较强的目标识别性能。

关键词: SAR图像, 特征选择, 目标识别和分类