Deep network based on up and down blocks using wavelet transform and successive multi-scale spatial attention for cloud detection

Cloud detection is not only a challenging task, but it also plays a major role in image processing. Due to the diversity of the cloud and the complexity of underlying surfaces, most of the current cloud detection methods still face great challenges, especially in detecting thin cloud. Therefore, we propose a method to detect cloud pixels in the GaoFen-1 WFV images. In our method, a deep network is used to learn the multi-scale global features so that the high-level semantic information obtained in the process of feature learning is integrated with the low-level spatial information in order to classify images into cloud and non-cloud regions. In addition, Up and Down blocks are designed using the Haar wavelet transform in order to fully exploit the structural information of images, and especially the texture information of the cloud, which can be learned targetedly. We pay attention to the original information of images in order to assist the learning for the network. In addition, we have also utilized dark channel prior and designed a Successive Multi-scale Spatial Attention Module by adding attention mechanisms to multi-scale feature maps in the network in order to provide a consistent improvement in the performance. The experimental results indicate that the proposed network performs well under different scenarios.

Automated summary

In this study, a deep network is proposed to detect cloud pixels in high-resolution satellite images, by learning multi-scale global features and utilizing Haar wavelet transform for structure and texture information. The method shows good performance in various scenarios, with consistent improvement achieved through a Successive Multi-scale Spatial Attention Module.