提出一种基于快速视网膜关键点(FREAK)局部不变特征的激光主动成像目标识别方法，在目标的矩、几何、轮廓等特征难以提取时仍然能够准确识别。提出差分加速分割检测特征(FAST-Difference)特征点检测方法，将加速分割检测特征(FAST)与高斯差分算子(DoG)相结合，快速提取出对于各种变换均具有稳健性的特征点；使用FREAK描述子生成特征向量，最后使用级联匹配方法完成特征向量的快速匹配。所提方法具有与人类视觉类似的特性：特征点检测阶段，FAST-Difference方法使用DoG进行多尺度检测；特征向量描述阶段，FREAK描述符的拓扑结构与人类视网膜类似；特征点匹配阶段，级联匹配方法模仿人类视觉的扫视搜索。实验证明所提方法能够对激光主动成像下的部分照亮的塔形目标进行快速有效的识别，算法满足识别系统实时性的要求。

A novel target recognition method for laser active imaging based on a kind of local invariant feature named fast retina key point (FREAK) is proposed, which could be used when moment, geometry and contour features of target are difficult to extract. A feature point detection method called difference of features from accelerated segment test (FAST-Difference) is proposed. It combines features from accelerated segment test (FAST) with difference of Gaussian (DoG), which can rapidly extract feature points that are robust for a variety of transformation. FREAK descriptor is used to generate the feature vectors and cascading matching method is used to complete the fast matching for feature vectors. The proposed method has characteristics similar to those of human vision. In the feature point detection stage, the FAST-Difference method uses DoG for multi-scale detection. In the feature vector description stage, FREAK descriptor has similar topology structure with human retina. In the feature vector matching stage, cascading matching method imitates saccade search of human retina. Experimental result indicates that the proposed method can quickly and effectively recognize the partly-illuminated tower-like target in laser active imaging and satisfy the real-time requirement of the recognition system.