Publications

Yin, MJ; Wang, P; Hao, WL; Ni, C (2023). Cloud detection of high-resolution remote sensing image based on improved U-Net. MULTIMEDIA TOOLS AND APPLICATIONS, 82(16), 25271-25288.

Abstract
Cloud detection is an important part of remote sensing image processing. Excellent cloud detection methods can accurately extract clouds from remote sensing images, providing convenience for subsequent processing. Most of the traditional cloud detection methods have problems such as poor generalization ability and insufficient segmentation accuracy, and the detection results cannot well meet the requirements of follow-up work. In this paper, the cloud detection dataset is constructed by data enhancement method, and the cloud in remote sensing image is accurately detected by improving U-Net network model. Firstly, the sub-pixel image segmentation is achieved by using the superpixels segmentation method to further improve the accuracy of similar target aggregation and reduce cloud labeling errors. Then, the Gaussian blur method is improved to blur the remote sensing image around the cloud label adaptively, which weakens the influence of background on cloud detection and reduces the computing cost. Finally, VGG16 network is used to deepen the feature extraction part of U-Net network to extract multi-scale cloud feature information from remote sensing images and improve the cloud detection accuracy. A large number of experiments are carried out on GF-2 and MODIS remote sensing images and compared with other cloud detection methods. Experimental results show that the proposed method can accurately detect large area clouds and broken clouds in remote sensing images, with Dice value up to 94.6%, IoU value up to 89.8% and CPA value up to 94.2%.

DOI:
10.1007/s11042-023-14655-z

ISSN:
1573-7721