一种改进的MRF点目标检测算法

摘要

针对复杂背景下点目标的单帧检测,明确提出有效像元的检测,基于点目标的局部相关性以及目标和背景的局部差异,提出了一种改进的基于马尔可夫随机场(Markov Random Field,MRF)的点目标检测算法.该算法依据一种基于复杂背景可分性度量的信杂比(Signal to Clutter Ratio,SCR)准则对MRF进行迭代优化的初始配置.在此基础上,改进了MRF标记场的先验概率模型,设计了一种基于欧式空间度量的MRF先验概率能量函数,构造了MRF对欧式空间距离的标记场概率响应模型,并通过高阶能量函数提高了目标概率对邻域标记变化的响应能力.分析结果表明:该算法在结构化背景中的性能更优,相比于传统Potts模型在目标辐射维度的检测能力更强,是一种鲁棒性更强的检测算法.%This paper focuses on point target detection with single frame under complicated background and suggests the conception of valid pixel detection.A modified point target detection method based on Markov Random Field was proposed in terms of local correlation of point target and local difference of target and background.This algorithm conducted initial configuration of iterative optimization for MRF by a signal-to-clutter ratio criterion based on complex background separability measure.Moreover,the prior probability model of MRF label field was improved by designing a new prior probability energy function based on Euclidean metric:firstly the label field probability response model of MRF to Euclid-ean space distance was buisecondly the response ability of the target probability to neighborhood la-bel change was improved by a higher order energy function.The results indicate that:the performance of the detection algorithm in structured background is better;the target's radiation-dimension detection ability of the modified label field prior probability model is more vigorous compared to the traditional Potts model.The proposed algorithm is a more robust one.