Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology, Nanjing University of Information Science and Technology
At present, the cloud detection based on high-resolution optical images, with deep learning methodology, cannot provide adequate and accurate information about the cloud, the cloud shadows or their edge details. The main reason lies in the insufficient fusion of semantic information in different scales of classification techniques. In response to this problem, this paper combines Res.block module that can prevent network degradation, the multi-scale convolution module (MCM) that can increase the receptive field of the network and the multi-scale feature module (MFM) that can extract and integrate information from different scales and proposes a detection algorithm based on Multi-scale Feature Fusion Network (MFFN) based on deep learning. The experimental results show that rich spatial information as well as semantic information can be extracted by the algorithm. Cloud and cloud shadow masks with higher level of accuracy can also be acquired. The accuracy of the cloud detection is 0.9351, and the accuracy of the cloud shadow detection is 0.8103. Meanwhile, the study provides theoretical support and technical reserve for the application of deep learning techniques on operational cloud detection.