几种典型红外弱小目标检测算法的性能评估

对基于中值相减滤波、最大中值相减滤波、最大均值相减滤波和推广的结构张量的红外弱小目标检测算法的性能进行了评估.针对传统评估方法的不足,提出了一种基于支持向量回归的红外弱小目标检测算法性能评估方法.利用该方法分别从图像背景特性和目标特性2方面对4种检测算法性能的影响进行定量分析和比较.实验结果表明,图像背景特性和目标特性对4种算法的检测性能都有较大的影响,而目标特性与4种算法的检测性能的依赖关系更明显;在4种评估算法中,基于推广的结构张量算法比其他3种传统红外弱小目标检测算法具有更好的鲁棒性.

Performance evaluation of several typical infrared weak and small target detection algorithms

Performances of infrared weak and small target detection (IWSTD) algorithms based on median and max-median subtraction filters, max mean subtraction filter and generalized structure tensor were evaluated in this paper. Firstly, According to the shortage of traditional methods, an IWSTD algorithm performance evaluation method based on support vector regression (SVR) was presented. Then, with this method, the influences of target characteristics and background characteristics over four algorithms were analyzed and compared quantitatively. Experiment results show that both target characteristics and background characteristics have significant effects on performances of four algorithms. Relatively, performance depends more obviously on target characteristics. And in four evaluated algorithms, the algorithm based on generalized structure tensor is more robust than other three traditional IWSTD algorithms.