国产高分辨率卫星影像云检测

Cloud Detection of Domestic High-resolution Satellite Imagery

针对高分辨率卫星影像在云检测过程中存在少云影像难以检测、云过渡区域检测不理想和类云地物难以去除的问题,提出一种联合阈值分割云检测算法和基于梯度的类云地物去除算法。首先,对影像进行梯度增强,以提高云与其他地物的光谱差异;其次,根据有云影像直方图呈现U形的特征,将影像分为有云和无云影像;然后,通过联合阈值分割方法检测云区域,并利用云区域内波谱特征验证云检测的正确性;最后,根据云与类云地物边缘梯度的差异性去除类云地物。选用高分一号、资源三号和天绘一号进行实验,并与大津法、树状结构法和目视解译结果进行对比。实验结果表明,所提出算法普适性强,不仅能够准确地检测多云影像,而且对于少云影像也具有较高的准确率,其检测效果优于大津法和树状结构法。

In view of the problems that partly cloudy image is difficult to detect,cloud transition area detection is not ideal and cloud-like objects are difficult to remove in the cloud detection process of high-resolution satellite images,this paper proposes a cloud detection algorithm based on joint threshold segmentation and a cloud-like object removal algorithm based on gradient. Firstly,the image is enhanced by gradient to improve the spectral difference between cloud and other ground objects. Secondly,according to the U-shaped feature of the histogram of cloud image,the image is divided into cloud image and no cloud image. Then,the cloud region is detected by the joint threshold segmentation method,and the correctness of cloud detection is verified by using the spectral characteristics in the cloud region. Finally,cloud-like objects are removed according to the difference of edge gradients between cloud and cloud-like objects. In this paper,GF-1,ZY-3 and mapping satellite-1 are selected for experiments,and the results are compared with those of Otsu method,tree structure method and visual interpretation. Experimental results show that the proposed algorithm has strong universality,which can not only accurately detect cloudy images,but also have higher accuracy for partly cloudy images,and its detection effect is better than that of Otsu method and tree structure method.