结合IR-MAD与均值漂移算法的密集城区遥感影像变化检测

不同时相遥感影像变化检测已成为土地利用变更调查、城市扩张分析、自然灾害分析及其他环境问题必不可少的技术手段之一。本文提出了一种结合IR-MAD与均值漂移算法的密集城区遥感影像变化检测方法。该方法通过伪不变特征法完成两期影像的相对辐射校正,有效改善影像间的配准误差,并利用IR-MAD算法对校正后的影像进行迭代运算,采用均值漂移算法对迭代后的影像进行分割,同时运用形态学方法处理分割后的影像,最终提取变化图斑。试验结果表明,该方法可以有效检测出变化区域,可应用于城市地表覆盖的变化检测。     

Relative Radiometric Normalisation - performance testing of selected techniques and impact analysis on vegetation and water bodies

In this paper, six image-based Relative Radiometric Normalization (RRN) techniques were applied to normalize the bi-temporal Landsat 5 TM data-set. RRN techniques do not require any atmospheric and ground information at the time of image acquisition. The target image for the year 2009 was normalized in such a way that it resembled the atmospheric and sensor conditions similar to those under which the reference image of the same season for the year 1990 was acquired. Among the selected methods applied, it was found that the Iteratively Reweighted Multivariate Alteration Detection (IR-MAD) method performed better, based on the error statistic. The IR-MAD technique was found to be advantageous as it identified a large set of true time-invariant pixels automatically from the change background using iterative canonical component analysis. The technique also stretches the values of Normalized Difference Vegetation Index and Normalized Difference Water Index and may help to distinguish different vegetation and water bodies better.     

多特征融合的高分辨率遥感影像飞机目标变化检测

为利用高分辨率遥感影像实现高精度的飞机目标变化检测,提出了一种自适应的多特征融合变化检测与深度学习相结合的方法。首先,通过加权迭代的多元变化检测法获取变化强度图,并结合自适应的直方图统计法自动获取显著的变化与不变化样本;然后,提取多时相影像的光谱、边缘和纹理特征,完成多特征融合的变化检测,并通过形态学处理得到变化图斑;最后,利用训练的NIN(Network in Network)结构的卷积神经网络飞机识别模型,完成变化图斑的类型判别,实现变化飞机的检测。实验结果表明,本文方法在两组数据的正确率分别达到100%和91.89%,均优于对比方法,能实现准确可靠的飞机目标变化检测。