摘要
当前红外目标检测算法因忽略了小目标的形状与温度信息,在杂波与传感器噪声干扰下,难以消除伪目标像素的干扰,导致其无法精确检测小目标,为此提出一种基于加权显著图与像素识别决策标准的红外小目标检测算法。基于目标形状与温度信息,引入自适应旷修剪均值滤波,建立加权显著图,依据其3D强度分布,将目标从复杂背景中凸显出来;基于目标像素的均值与标准偏差,定义阈值决策模型,确定参考目标区域及其中心点;通过不断更新阀值,联合参考目标区域像素与待检测像素的紧密度与相似度,构建目标识别策略,从复杂背景中检测出完整的目标。实验结果表明,与当前红外目标检测技术相比,在严重噪声干扰下,该算法检测精度更高,有效剔除了伪目标像素,具有更高的信噪比收益与效率,表现出更好的ROCs特性曲线。
To solve the defect of low target detection accuracy induced by difficulties to resist the interference of the sensor noise and clutter in current infrared target detection algorithm,the infrared small target detection algorithm based on weighted saliency map and pixel recognition decision criteria was proposed.The weighted saliency map was built by introducing the a-trimmed mean filter and target shape,as well as temperature information to enhance the small target,and the target was highlighted from the complex background according to its intensity distribution.The threshold decision model was constructed using the mean and standard deviation of the target pixel to determine the reference target region and its center point.Through continually updating the threshold,the target identification standard was built according to the adjacency and similarity of reference target pixels and pixel detection to detect the complete target from the complex background.Experimental results show that this algorithm has higher detection accuracy to eliminate the noise points with higher signal to clutter ratio and better ROCs curve under the condition of serious noise interference compared with the current infrared target detection technology.
出处
《计算机工程与设计》
北大核心
2017年第7期1898-1904,共7页
Computer Engineering and Design
基金
国家自然科学基金项目(61379079)
河南省国际科技合作基金项目(144300510007)
关键词
红外目标检测
加权显著图
α-修剪均值滤波
阈值决策
紧密度
目标识别策略
infrared target detection
weighted saliency map
alpha trimmed mean filter
threshold decision
adjacency
target identification standard
