目标检测性能的定量评估方法

Quantitative performance evaluation method for target detection

提出了一种对目标检测算法性能进行定量评估的方法。该方法基于ROC曲线定义了不同检测算法在Gaussian分布假设下的等效输入信杂比(EISCR),并用该EISCR的大小来对不同检测算法的性能进行定量评估。提出用Sigmoid函数对Gaussian分布函数逼近的方法推导了EISCR的简化计算公式。该方法克服了以往性能评估方法不能对检测算法的性能进行定量评估的缺点,不仅可以评价不同CFAR算法的性能,还能定量评估CFAR算法与非CFAR算法之间的性能差异。仿真结果表明了该方法的有效性。

A quantitative performance evaluation method for target detection is propcxsed. Based on the Receiver Operating Characteristic （ROC） curve, the Equivalent Input SCR （EISCR） is defined under Gaussian distribution assumption. The performance of a detection method is evaluated quantitatively with the value of EISCR. By approximating Sigrnoid function to Gaussian distribution function, a concise formula for EISCR calculation is deduced. With this method the performances of different detection methods can be evaluated＇quantitatively, thus overcoming the shortcoming of the existing evaluation methods. The method can be used to evaluate not only the performances of different CFAR methods, but also the performance difference between CFAR methods and non-CFAR methods. Simulation results show the validity of the method.