基于最小二乘支持向量机时域背景预测红外弱小目标检测.pdfVIP

31 6 Vol. 31 No. 6 2 0 1 0 6 ACTA ARMAMENTARII Jun. 2010 基于最小二乘支持向量机时域背景 预测的红外弱小目标检测 吴一全, 罗子娟 ( , 210016) : 针对信噪比较低时, 如何有效地抑制自然背景对目标检测的影响, 提出了一种 于最小 二乘支持向量机(LS-SVM )时域背景预测的红外弱小目标检测方法首先针对前几帧图像中对应 同一位置像素点的灰度值序列, 利用参数经粒子群优化的最小二乘支持向量机进行函数拟合, 并据 此预测下 一帧图像在该位置处像素点的灰度值; 然后将原始图像与预测图像相减得到预测残差图 像, 利用 于二维Tsallis- avrda-Charvat熵的阈值选取快速算法进行分割, 并根据小目标运动的连 续性和轨迹的 一致性进 一步分离噪声和小目标文中给出了实验结果及分析, 并与现有的检测红 外小目标的空域和时域背景预测算法进行了比较结果表明所提出的算法具有更高的检测概率, 明显优于已有的 于背景预测的红外小目标检测算法 : 信息处理技术; 红外小目标检测; 背景预测; 最小二乘支持向量机; 粒子群优化; 二 维Tsallis- avrda-Charvat熵 : TN91173; TP3914 : A : 1000-1093(2010)06-0678-07 Small Infrared TargetDetection Based on Least Squares Support VectorM achine TemporalBackground Prediction WU Yi-quan, LUO Z-ijuan ( Schoolof Information Science andTechnology, NanjingUniversity ofAeronautics and Astronautics, Nanjing 210016, Jiangsu, China) Abstract: A mi ing at suppressing the influence ofnaturalbackground on the targetdetection effectively in lower signa-l to-noise ratio, amethod of small target detection in infrared mi age sequencewas proposed based on the least squares support vectormachine (LS-SVM ) temporalbackground predication. Firstly, the gray value sequences at the same pixel locations in the previous frameswere fitted by using the LS- SVM optmi ized by the particle swarm. The gray value at the same pixel location in the next frame could be predicted by the fitted function. Then, the estmi ated mi age subtracted from the source mi age gave the residual mi age. The residual mi agewas segmented by using two-dmi ensionalTsallis-