变化检测是指利用多时相特征检测地表覆盖类型发生变化的区域,目前的检测方法易受噪声以及特殊地物等影响,使得检测结果斑点现象严重,检测精度低。针对这一问题,本文结合典型相关分析和直方图规定化提出一种非监督的超像素级变化检测方法。首先,对两时刻的遥感影像进行预处理以及超像素分割；其次,基于S-IR-MAD方法和超像素大小等因素计算每个超像素的权重；然后以超像素为分析对象基于典型相关分析(CCA)计算特征影像,并对特征影像进行直方图规定化来减弱特殊地物的影响；最后,结合权重影像与检测结果进行决策融合。本文在三个高光谱测试数据集和一个多光谱测试数据集上进行实验验证。结果表明,本文方法在四个测试数据集上的整体检测精度均为最优,且都达到了90%以上,在Barbara数据集和Bay Area数据集上的提升效果最为明显,精度分别提高了3.43%和5.35%,对噪声具有良好的鲁棒性,对实际应用具有一定的参考价值。

Change detection is the process of detecting the changed surface cover using the multi-phase characteristics. The current detection method is easy to be affected by noise and special objects, which makes the detection result of spot phenomenon serious and the detection accuracy low. To solve this problem, this paper proposes an unsupervised super pixel level change detection method based on canonical correlation analysis and histogram specification. Firstly, we preprocess two remote sensing images and segment the images with super pixel. Secondly, we calculate the weight of super pixel based on S-IR-MAD and the size of super pixel. What’s more, the feature image is calculated based on canonical correlation analysis (CCA) taking super pixel as analysis object. To reduce the influence of special objects, histogram specification is used to feature image. Finally, we do decision fusion based on weight image and detection result.