Deep learning based cloud detection for medium and high resolution remote sensing images of different sensors (1810.05801v3)

Abstract: Cloud detection is an important preprocessing step for the precise application of optical satellite imagery. In this paper, we propose a deep learning based cloud detection method named multi-scale convolutional feature fusion (MSCFF) for remote sensing images of different sensors. In the network architecture of MSCFF, the symmetric encoder-decoder module, which provides both local and global context by densifying feature maps with trainable convolutional filter banks, is utilized to extract multi-scale and high-level spatial features. The feature maps of multiple scales are then up-sampled and concatenated, and a novel multi-scale feature fusion module is designed to fuse the features of different scales for the output. The two output feature maps of the network are cloud and cloud shadow maps, which are in turn fed to binary classifiers outside the model to obtain the final cloud and cloud shadow mask. The MSCFF method was validated on hundreds of globally distributed optical satellite images, with spatial resolutions ranging from 0.5 to 50 m, including Landsat-5/7/8, Gaofen-1/2/4, Sentinel-2, Ziyuan-3, CBERS-04, Huanjing-1, and collected high-resolution images exported from Google Earth. The experimental results show that MSCFF achieves a higher accuracy than the traditional rule-based cloud detection methods and the state-of-the-art deep learning models, especially in bright surface covered areas. The effectiveness of MSCFF means that it has great promise for the practical application of cloud detection for multiple types of medium and high-resolution remote sensing images. Our established global high-resolution cloud detection validation dataset has been made available online.

Overview of a Deep Learning-Based Cloud Detection Framework in Remote Sensing

This paper introduces a novel deep learning method for cloud detection in satellite imagery, known as Multi-Scale Convolutional Feature Fusion (MSCFF). Cloud detection is a critical preprocessing task in remote sensing, as clouds can obscure the ground information in optical satellite images. The MSCFF method addresses the limitations of previous techniques by providing robust cloud and cloud shadow detection across multiple satellite sensors and resolutions, ranging from 0.5 to 50 meters.

In traditional cloud detection approaches, rule-based methods utilizing the physical properties of clouds have been prevalent. These methods, while effective in certain scenarios, often struggle with high false positives in bright surface areas and thin cloud omission. Deep learning techniques have emerged as a promising alternative, with architectures like CNNs showing potential in this domain. Previous studies, however, have often been constrained to specific regions and types of imagery.

Methodology and Implementation

The MSCFF method is designed with a fully convolutional network architecture leveraging a symmetric encoder-decoder module for feature extraction and a novel multi-scale feature fusion process. The method takes advantage of the encoder-decoder architecture to retain both low-level spatial details and high-level semantic features, enhanced by residual network units and dilated convolutions. This architecture allows MSCFF to effectively discriminate clouds from non-cloud bright objects. The output of the MSCFF network consists of two maps, indicating clouds and cloud shadows, respectively, which are further processed through binary classifiers for final mask generation.

The training of MSCFF employs a comprehensive global high-resolution cloud detection validation dataset, consisting of various satellite images, including Landsat-7/8, Gaofen-1/2/4, Sentinel-2, and others. These datasets encompass a wide range of geographical locations and land-cover types, improving the generalization capabilities of the model.

Results and Comparisons

The effectiveness of MSCFF is validated against existing rule-based and deep learning methods, such as Fmask, MFC, PRS, DeepLab, and DCN. The MSCFF model consistently delivers higher accuracy scores, particularly in environments with bright surfaces, such as snow or urban areas. For example, the MSCFF method improves the mean Intersection over Union (mIoU) scores compared to traditional methods while reducing false positives associated with bright non-cloud surfaces.

Implications and Future Directions

The MSCFF method shows great promise for practical applications in remote sensing, enabling efficient preprocessing of diverse satellite imagery. In terms of both theoretical and practical implications, MSCFF contributes to the advancement of deep learning applications in remote sensing, encouraging further exploration of adaptable cloud detection methods across various sensor types and resolutions.

Future research directions could explore the enhancement of cloud shadow detection, leveraging techniques like object-based image analysis to address challenges of scale and shadow-cloud association in high-resolution imagery. Expanding the application of MSCFF to additional types of satellite data and improving the robustness of training strategies for underrepresented classes within the datasets represent additional areas for development.

Overall, this paper advances the field by introducing a robust, sensor-agnostic cloud detection framework, highlighting the potential of deep learning in overcoming the limitations of traditional methods in remote sensing applications.