Gated aggregation network for cloud detection in remote sensing image

Cloud detection is one of the important tasks in remote sensing image preprocessing, and this paper uses RGB preview images of remote sensing images to extract cloud regions efficiently. The preview images make the detection of cloud regions more challenging due to the lack of resolution and spectra...

Full description

Saved in:
Bibliographic Details
Published in:The Visual computer 2024-04, Vol.40 (4), p.2517-2536
Main Authors: Du, Xianjun, Wu, Hailei
Format: Article
Language:English
Subjects:
Algorithms
Artificial Intelligence
Clouds
Computer Graphics
Computer Science
Datasets
Deep learning
Design
Image Processing and Computer Vision
Modules
Neural networks
Original Article
Performance evaluation
Remote sensing
Satellites
Semantics
Spatial resolution
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Cloud detection is one of the important tasks in remote sensing image preprocessing, and this paper uses RGB preview images of remote sensing images to extract cloud regions efficiently. The preview images make the detection of cloud regions more challenging due to the lack of resolution and spectral information. The existing remote sensing image cloud detection methods, and feature fusion process, due to unreasonable feature fusion strategy, so that the encoded features cannot be fully utilized, which may introduce noise information, and ultimately lead to the problems of false detection and missing detection. To address these problems, this work designs a gated aggregation network (GANet) for remote sensing image cloud detection. GANet has a novel encoder–decoder architecture, a gated feature aggregation module (GFAM), and a pyramidal attention pooling module (PAPM). GFAM bridges the gap between high resolution with spatial details and low-resolution features with high-level semantics, fully selectively fusing semantic and spatial features to alleviate the semantic divide problem when fusing multi-level features. PAPM extracts multi-scale global contextual features without loss of resolution. The method is validated on three datasets: the publicly available 38-Cloud and SPARCS datasets and the self-built Landsat-8 cloud detection dataset with higher spatial resolution. The experimental results show that the proposed method achieves competitive performance under different evaluation metrics. Codes and datasets can be found at https://github.com/HaiLei-Fly/GANet/ .
ISSN:0178-2789
1432-2315
DOI:10.1007/s00371-023-02934-7