YOLO Based Cloud Region Recognition Algorithm for Remote Sensing Images*

Remote sensing imaging is able to acquire huge amount of valuable information from the ground to support the development of various technologies. However, clouds are the obstacles mostly met which lead to degraded images. Studying on removing cloud from single remote sensing images is significant to enhance the image information underneath clouds. Combining the operations of the cloud region recognition and the cloud removal is expected to remove clouds effectively and avoid the losing of image information outside of the clouds as much as possible. In this paper, inspired by object detection technology using the deep learning networks, an algorithm of recognizing cloud regions from single images based on YOLO was proposed. The feasibility of cloud region recognition was discussed after the characteristics of remote sensing images disturbed by clouds were analyzed. A simulation method was developed to produce image sets on the basis of the atmosphere transmission model of remote sensing imaging. Experiments were carried out by applying YOLOv5 and YOLOv8 to recognize the cloud regions in remote sensing image. The network with several weight pre-training models of YOLOv5 and YOLOv8 was trained and tested respectively with the simulation image sets. Results of cloud region recognition prove that YOLOv5-l achieved the best successful rate to recognize clouds from single images which reached over 95%.