摘要
声纳图像中人造目标的自动检测是当前水下探测领域需要重点解决的问题之一。传统的基于目标回波信号强度的检测方法在海底存在岩石等类似于水雷等人造目标的情况下,常会导致较高的虚警率。由于人造目标和自然背景之间的纹理特性的不同,自然背景一般具有较复杂的纹理,而人造目标形状规则、表面光滑、纹理简单。利用分形模型中分形维数特征、截距特征、分形拟合误差特征以及多尺度分形特征进行声纳图像人造目标检测,仿真实验表明基于分形的检测算法可较好地实现人造目标和自然背景的分离,从而为成像声纳水下探测技术的进一步发展奠定了基础。
The detection of man-made object in sonar image is one of crucial problems in the field of underwater detection.Conventional methods based on the intensity of target echo signal will result in high false alarming rate if mine-like object such as rock laying on the seafloor.For the texture difference between man-made object and natural background,natural background has always complicated texture and man-made object has simple texture and regular shape.The fractal dimension,intercept,polynomial error and multi-scale fractal feature are adopted to detect man-made object in sonar image.Simulation resuhs are presented and indicate that the detection performance of this algorithm is satisfied,which provides proper condition for underwater detection and recognition.
出处
《计算机工程与应用》
CSCD
北大核心
2006年第36期195-197,211,共4页
Computer Engineering and Applications
关键词
人造目标
分形特征
声纳图像
纹理
man-made object
fractal feature
sonar image
texture