针对红外目标识别问题，提出了一种基于协方差描述子和核稀疏编码的红外目标识别方法。该方法结合了红外图像的灰度、一阶以及二阶梯度等特征的协方差描述子作为红外目标的特征，并采用 Log-Euclidean度量进行特征相似性计算，通过高斯核函数将协方差描述子映射到高维空间，最后在新的特征空间上进行稀疏编码。实测数据实验结果表明，与传统的 KNN（k-nearest neighbor，k最近邻）以及 SVM（support vector machine，支持向量机）等分类算法相比，基于核稀疏编码的红外识别方法在识别准确率上有很大的提高。

An infrared target recognition method based on covariance descriptor and kernel sparse coding isproposed in this paper. Covariance descriptor combining infrared image gray intensity values and the normof first and second order derivatives of the intensities as infrared image features. With respect to x and y isextracted as feature representation, similarity of covariance descriptors is computed through Log-Euclideanmetric. Then covariance descriptors are mapped into a high dimensional feature space through Gaussiankernel function. Finally, infrared target recognition is accomplished using sparse coding in the new featurespace. Experiments on the real infrared images show that our method obtains better results compared withtraditional algorithms such as KNN and SVM.