多维特征点空间的红外弱小目标检测方法  

Multi-dimensional feature point space infrared dim target detection method

作　　者：王芳[1] 王海晏[1] 寇添 聂光戍[1] WANG Fang;WANG Haiyan;KOU Tian;NIE Guangshu(College of Aeronautics Engineering,Air Force Engineering University,Xi’an 710038,China)

机构地区：[1]空军工程大学航空工程学院,陕西西安710038

出　　处：《应用光学》2020年第6期1268-1276,共9页Journal of Applied Optics

基　　金：国家自然科学基金(61172038)。

摘　　要：将人工智能算法引入目标检测,空间红外弱小目标的检测也可归为模糊检测的二分类问题。依据空中红外弱小目标的探测模型,建立了信号电压比光谱模型,仿真分析表明电压比变化趋势与目标的速度、姿态和两机态势有关,可用以检测目标。采用动态特征构建理论,构建了红外弱小目标的双色比特征空间,基于该特征空间,优化最小二乘分类算法,用于从光谱信号层级检测目标。该方法不仅缩小了样本数据量,而且防止了高斯核函数参数选择引起的"过拟合"现象,既保证了分类精度,又使分类速率提高近1倍,为人工智能算法用于红外弱小目标检测提供了参考依据。As artificial intelligence algorithm was introduced into target detection,the detection of spatial infrared dim targets could be classified as the binary problem of fuzzy detection.According to the detection model of infrared dim target in the air,a signal voltage ratio spectrum model was established.The simulation analysis showed that the variation trend of voltage ratio was related to the speed,attitude of the target and the two-machine posture,which could be used to detect the target.The dynamic characteristics building theory was adopted to build the bicolor ratio feature space of infrared dim target.Based on this feature space,the least squares classification algorithm was optimized to identify the objects from the spectral signal hierarchy.This method not only reduces the amount of the sample data,but also prevents the phenomenon of over-fitting caused by the parameter selection of Gaussian kernel function.It ensures the classification accuracy and improves the classification efficiency nearly doubled.Reference basis is provided for infrared dim target detection by artificial intelligence algorithm.

关 键 词：多光谱探测 双色比特征空间 弱小目标检测 最小二乘分类算法 

分 类 号：TN215[电子电信—物理电子学]