Abstract:Cloud detection is a prerequisite for optical remote sensing applications due to the ubiquitous cloud coverage and negative effect of cloud occlusions. However, it is very challenging because of the heterogeneity of clouds and diversity of underlying surfaces, especially for thin clouds with illegible shapes and dispersed distribution in addition to high transparency, which is a common bottleneck for most methods. To tackle these problems, this article proposes an innovative attention-aided multilevel cloud detection network (AMCD-Net) for optical satellite imagery. Specifically, AMCD-Net takes full account of the variability and complexity of clouds and integrates multilevel features and different attentions within a deep convolutional U-Net framework, thus enabling more accurate cloud identification in complex scenarios. On the one hand, a multilevel asymmetric convolutional (MAC) module embedded encoder is established to learn discriminative representations for clouds with various shapes and integrate the complementarity of multilevel features to improve model robustness, with a regional attention-based decoder constructed to more accurately recover complicated cloud distribution, which effectively balances the integrity and details of clouds. On the other hand, a deformable convolution-based geometry enhancement attention (GEA) is designed to refine information transmission between the encoder and the decoder, with a joint loss of binary cross-entropy (BCE) and structural similarity index (SSIM) constructed to simultaneously focus on category and morphology discriminant excavation, which are favorable for fine-grained cloud prediction. The effectiveness of AMCD-Net was verified on two well-known datasets, i.e., the 38-Cloud dataset and the SPARCS dataset, and the results demonstrate that it outperformed the other state-of-the-art deep networks.

Auroral Image Classification Based on Second-Order Convolutional Network and Channel Attention Awareness

Multiview Learning for Automatic Classification of Multiwavelength Auroral Images

Multi-view learning for automatic classification of multi-wavelength auroral images

A Hyperspectral Image Classification Method Based on the Nonlocal Attention Mechanism of a Multiscale Convolutional Neural Network.

Automated Classification of Auroral Images with Deep Neural Networks

Aerial Scene Classification Via Multilevel Fusion Based on Deep Convolutional Neural Networks.

Detection of Diffuse Auroral Events Based on LSTM

A new ART neural networks for remote sensing image classification

Extracting Auroral Key Local Structures from All‐Sky Auroral Images by Artificial Intelligence Technique

An appearance quality classification method for Auricularia auricula based on deep learning

A3 CLNN: Spatial, Spectral and Multiscale Attention ConvLSTM Neural Network for Multisource Remote Sensing Data Classification

An integrated aurora image retrieval system: AuroraEye

Group-Wise Learning for Aurora Image Classification With Multiple Representations

Auroral breakup detection in all-sky images by unsupervised learning

A Novel Multi-scale Attention Feature Extraction Block for Aerial Remote Sensing Image Classification

Enhanced Multi-level Features for Very High Resolution Remote Sensing Scene Classification

AMCD-Net: An Effective Attention-Aided Multilevel Cloud Detection Network for Optical Satellite Imagery

ReSC-net: Hyperspectral Image Classification Based on Attention-Enhanced Residual Module and Spatial-Channel Attention

Channel-Attention-Based DenseNet Network for Remote Sensing Image Scene Classification

Hyperspectral Image Classification Based on Atrous Convolution Channel Attention-Aided Dense Convolutional Neural Network

Cloud Detection Method Using CNN Based on Cascaded Feature Attention and Channel Attention