APP下载

基于频谱分布量化分析的星载人工目标快速检测方法

2017-10-17李文娟赵和平庞波

现代电子技术 2017年20期

李文娟 赵和平 庞波

摘 要: 地物特征不单靠空域中像元灰度值的变化程度来表现,其频谱能够表征不同地物的特征,并可通过该频谱特征进行遥感图像的分析和目标的搜索、检测。提出一种通过频谱分析快速发现自然背景中人工目标及其所在区域的新方法,首先对图像的频谱分布特征进行了分析并提出了一种频谱分布量化的算法,定义频谱分布量化指数;根据计算出的区域频谱分布量化指数,借鉴视觉显著性的思想计算出各区域在整幅图像中的显著值,从而实现人工目标的检测。文中对该方法在海背景中舰船目标的检测和沙漠、戈壁中人工建筑的检测进行了实验,从结果可以看出新算法能够有效地实现自然背景潜在人工目标的快速搜索和发现,并确定出其所在区域,具有一定的灵活性,能够适应星上多变的检测场景。

关键词: 图像处理; 傅里叶变换; 频谱分析; 显著检测

中图分类号: TN911.73?34; TP391 文献标识码: A 文章编号: 1004?373X(2017)20?0139?04

Abstract: Ground object features can be reflected not only by pixel gray value changes in spatial domain, but also by their spectrums, whose features can be used for remote sensing image analysis, target search, and target detection. Therefore, a new fast detection method of man?made targets in natural background and their locations by using spectrum analysis is proposed. First, the spectrum distribution characteristics of the image are analyzed, and then a spectrum distribution quantification algorithm is put forward to define the quantification index of spectrum distribution. According to the calculated quantification index of spectrum distribution in the region, the salience value of each region in the whole image is calculated by drawing on the idea of visual saliency, to realize the detection of man?made targets. In this paper, an experiment of this method to detect ship targets in the sea background and man?made structures in desert and gobi is carried out. The results show that the new algorithm can effectively realize the rapid search and discovery of potential man?made targets in natural background, determine their locations, has certain flexibility, and can adapt itself to the changeable detection scenarios on satellites.

Keywords: image processing; Fourier transform; spectrum analysis; saliency detection

0 引 言

在星上資源和时间受限的条件下,如何通过遥感图像信息处理使卫星能够快速聚焦到目标或者目标区域是星载遥感图像信息处理的重点问题之一。目标检测是星上图像信息处理的重要研究领域,其通常算法是对图像进行分割,提取出目标的特征,再通过分类器根据特征实现人工目标的检测。这类算法通常可分为两种:一种是由下而上的数据驱动方法,另一种是由上而下的知识驱动方法[1]。……

登录APP查看全文