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

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,提出了一种基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,利用基于二维Tsallis-Havrda-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,明显优于已有的基于背景预测的红外小目标检测算法。

Small Infrared Target Detection Based on Least Squares Support Vector Machine Temporal Background Prediction

Aiming at suppressing the influence of natural background on the target detection effectively in lower signal-to-noise ratio,a method of small target detection in infrared image sequence was proposed based on the least squares support vector machine(LS-SVM) temporal background predication.Firstly,the gray value sequences at the same pixel locations in the previous frames were fitted by using the LS-SVM optimized by the particle swarm.The gray value at the same pixel location in the next frame could be predi...