针对空间目标红外识别中典型的类内变化大、类间变化小等问题, 提出了一种用支持向量机(Support Vector Machine, SVM)分类器结合局部二值模式(Local Binary Pattern, LBP)直方图特征和灰度直方图特征的空间目标红外特征提取与识别方法。以国内某卫星和国外某卫星为研究对象, 提取它们红外图像的 LBP直方图特征以及灰度直方图特征; 使用红外仿真软件生成两个目标在不同姿态、不同分辨率下的样本图, 并分成两部分, 分别作为 SVM分类器的训练集和测试集。实验结果表明, LBP直方图特征和灰度直方图特征均能够以较高的准确率对空间目标进行识别, 且其识别效能与目标红外图像的分辨率以及 SVM核函数有关。

Dissimilarity is usually large among the same space target but small among different ones for infrared recognition. Aiming at these problems, a method of space-target infrared feature extraction and recognition with the use of SVM classifier combining with LBP histogram feature and gray histogram feature is proposed. A domestic satellite and a foreign satellite are taken as the objects of study. LBP histogram feature and gray histogram feature of their infrared images are extracted. Sample images under different attitudes and resolutions are generated by the simulation software, and they are divided into two parts, one for SVM classifier training set and the other for test set. The test results indicate that not only can both LBP histogram feature and gray histogram feature recognize the space targets with a high accuracy, but also their recognition performances relate to the resolution of infrared images and the kernel functions of SVM.