CLOUD DETECTION OF REMOTE SENSING IMAGES BY DEEP LEARNING
Mengyun Shi, Fengying Xie, Yue Zi, Jihao Yin
Image Processing Center,
School of Astronautics, Beihang University
Beijing 100191, P.R. China
ABSTRACT
Cloud detection plays a major role for remote sensing image
processing. Most of the existed cloud detection methods use
the low-level feature of the cloud, which often cause error
result especially for thin cloud and complex scene. In this
paper, a novel cloud detection method based on deep learning
framework is proposed. The designed deep Convolutional
Neural Networks (CNNs) consists of four convolutional
layers and two fully-connected layers, which can mine the
deep features of cloud. The image is firstly clustered into
superpixels as sub-region through simple linear iterative
cluster (SLIC) method. Through the designed network model,
the probability of each superpixel that belongs to cloud region
is predicted, so that the cloud probability map of the image is
generated. Lastly, the cloud region is obtained according to
the gradient of the cloud map. Through the proposed method,
both thin cloud and thick cloud can be detected well, and the
result is insensitive to complex scene. Experimental results
indicate that the proposed method is more robust and
effective than compared methods.
Index Terms—Cloud Detection, Convolutional Neural
Networks, Deep Learning, Superpixel
1. INTRODUCTION
With the rapid development of remote sensing images
acquisition technology, remote sensing images are widely
applied to various fields such as reconnoiter measure,
geographical mapping, resource monitoring. However, there
are often cloud regions occurrence in remote sensing images
because of the weather factor. These cloud regions will not
only cause loss of collected information but also make much
difficulty to target detection, object recognition and other
further processing tasks so as to generate wrong results in
subsequent analysis. Hence, detecting and removing cloud
regions from the images can improve the effectiveness of
remote sensing image interpretation. In the years, a number
of cloud detection methods have been proposed. Early
methods are designed in low spatial resolution images such
as NOAA/AVHRR images with about 1 km
2
/pixel [1].
Nevertheless, cloud detection of high resolution remote
sensing images has become a hot research topic in recent
years. Li et al. [2] use the brightness characteristics and
texture features to detect cloud from the remote sensing
image based on Support Vector Machine. Zhang and Xiao [3]
proposed a progressive refinement scheme for detecting
clouds in the color aerial photographs. An and Shi [4]
proposed an automatic supervised approach based on the
“scene-learning” scheme and regarded the cloud detection
problem as a classification task. In remote sensing images,
ground objects and cover are diverse, and cloud thickness and
form are varied. While most of existed cloud detection
methods extract low-level features, they hardly adapt to the
complex remote sensing images, especially for thin cloud in
low contrast background.
In this paper, a novel cloud detection method based on
deep learning is proposed for remote sensing images. Firstly
SLIC [5] method is used to cluster the image into small
superpixels, and for each superpixel, CNNs is used to learn
the feature of the cloud to generate the cloud probability
image. Through the gradient of the probability map, the cloud
is detected from the image. The rest of the paper is organized
as follows. Section II introduces the proposed cloud detection
method. Experiment is presented in Section III. And Section
IV summarizes the contributions.
2. OUR APPROACH
Deep learning methods have obtained great successes for a
wide range of tasks in computer vision, such as image
classification [6], object detection [7]. In this section, we
present the details for our cloud detection method based on
deep learning. Fig.1 shows the flowchart of the proposed
method. Firstly, image is clustered into superpixels using
SLIC method. Then, cloud probability map is generated using
designed CNNs. Finally, cloud region is detected through
refining the cloud probability map.
2.1. Obtaining superpixel sub-regions
Generally, cloud detection methods by machine learning take
each pixel as a unit, which is time-consuming and will
produce fragmented noise. In this paper, SLIC is used to
divide remote sensing image into superpixels and we use
superpixel as basic unit in the following steps.
SLIC performs a local clustering of pixels and can
generate compact superpixels as sub-regions, which adhere
well to region boundaries. According to [5], the clustering is
701978-1-5090-3332-4/16/$31.00 ©2016 IEEE IGARSS 2016