提出了一种用于扫描型红外预警系统的目标检测与识别算法来实现对空中威胁的预警,该算法在对目标进行检测和识别的过程中分别运用了降维技术。首先对拉普拉斯-高斯(Laplacian of Gaussian, LOG)算子降维, 改变经典LOG算子各向同性的特点, 从而减少了信息的丢失; 然后通过设置相关参数设计8邻域方位滤波器对图像不同方向进行尺度优化的滤波以完成目标增强与背景抑制; 最后对滤波结果进行目标提取, 通过支持向量机算法对目标进行识别。为使识别过程更加简捷而不失准确性, 在识别前采用基于协方差算子的充分降维方法对样本特征和目标特征进行降维, 从而在简化经典滤波算法与目标识别算法的同时提升了算法效率。实验结果表明, 与经典高维算法相比, 本文提出的算法在对红外目标进行检测识别时能够获得更好的效果, 应用于工程时能实现低于7%的虚警率和低于5%的漏警率, 且算法能够满足系统实时性要求。

An algorithm combined target detection and target recognition for a scanning infrared early-warning system was proposed to realize early-warning of air threats. First, the dimension of Laplacian of Gaussian (LOG) operator was reduced to change the isotropic characteristics of classical LOG operator and to reduce its information losses. Then, an eight-neighborhood local filter was proposed to enhance targets and suppress backgrounds by setting relevant parameters. Finally, the targets were extracted from filted results and were recognized by the Support Vector Machine(SVM) algorithm. In order to simplify the recognition procedure within preciseness, the Sufficient Dimension Reduction (SDR) based on the covariance operator was used to reduce the feature dimensions of samples and targets before recognition so that to simplify the classical filter and recognition algorithm while to improve the algorithm efficiency. Experimental results indicate that the proposed method gets better results than the high-dimensional algorithm and it can satisfy the system requirements of real-time performance. The false-warning rate and the miss-warning rate are lower than 7% and 5%, respectively.