Cloud detection plays an important role in remote sensing image quality evaluation, information acquisition, and analysis. For most optical satellites, the acquired multispectral bands include only three visible bands and one near-infrared band. How to achieve fast and accurate cloud detection with this limited number of bands is a problem worth studying. In this paper, a cloud detection algorithm (SMFCD) based on a saliency-based mixed feature map of images is proposed. This algorithm first calculates the saliency characteristics from the band mean and haze optimized transformation (HOT) map. The transmission map obtained from dark channel prior theory is used to combine the mixed feature map for subsequent detection. The proposed algorithm generates the initial thick cloud mask using a segmentation method based on the Otsu algorithm later. Guided filtering is then used to refine the cloud mask. The final cloud detection result is acquired through post processing. The algorithm is tested on six different datasets. It is shown that the algorithm can obtain good detection results for most images, with overall accuracies from 0.860 to 0.969 on these datasets. The average time consumption on these datasets reaches 4.4~42.1 seconds, and a fast version of the proposed algorithm reduces these times by about 50%. The proposed algorithm can be used for remote sensing image quality evaluation and subsequent application preprocessing with greater adaptability and flexibility than existing algorithms, since it does not depend on specific satellite radiometric calibration coefficients.

Cloud detection is one of the crucial tasks in the field of remote sensing, which is regarded as binary classification at the pixel level. Although a large number of recent deep learning-based methods have made great progress, most of their appealing performances come at the expense of a large amount of computation, which reduces the real-time performance accordingly. In order to bridge the gap between segmentation performance and inference speed, we propose a novel asymmetric nested network architecture termed ANNet, which is designed for real-time cloud detection with excellent performance. In the encoder branch of ANNet, we introduce an effective Tiny U-shape Block (TUB) to enrich detailed spatial contexts in each stage, which also allows ANNet to embed spatial recovery ability earlier, and a lightweight Simple Feature Fusing Module (SFFM) is designed to refine the semantic features map at a lower level of TUB for better performance. Following the encoder, which is diverse from the most symmetric U-shape approaches, an Asymmetric Light Decoder (ALD) with only convolution and bilinear up-sample operations is employed for spatial recovery. Moreover, we also demonstrated that utilizing a constant channel size instead of a larger channel volume as the network goes deeper is an efficient and effective design for cloud detection tasks. Substantial experiments are performed on GF1_WHU and 95-Cloud datasets, which show that ANNet has achieved the excellent performance with low computation cost compared to most of the existing state-of-the-art methods. On GF1_WHU dataset, ANNet-l achieves 93.79% on mIoU at 125 FPS, while ANNet-s with only 29.5K parameters yields 90.74% mIoU at 251 FPS on the Nvidia RTX 2080Ti.

Super-resolution reconstruction for optical remote sensing images enhances the spatial resolution of satellite imagery and video, improving detection and identification accuracy. This technology is crucial in fields such as public service, disaster response, national defense, and military applications. Dynamic target detection, particularly in intelligent transportation and military remote sensing, has attracted significant attention. Extracting motion information from dynamic targets aids in studying their behavior and enables tracking and trajectory prediction. However, factors like environmental conditions and hardware limitations often degrade the quality of remote sensing video, leading to loss of detail in dynamic targets. Restoring high-resolution dynamic targets from satellite video sequences remains a challenge. In this paper, we use satellite video data from the Luojia3-01 satellite to address this issue. First, we construct a super-resolution reconstruction dataset from pre-processed satellite video, containing multiple dynamic targets. We then propose a new video super-resolution network, ETDM_Plus, which incorporates a dynamic fusion attention module to enhance resolution by focusing on dynamic targets and utilizing spatiotemporal information. Experimental results show that in the dynamic target area, the proposed method improves peak signal-to-noise ratio (PSNR) by 2.28%–3.85% and structural similarity (SSIM) by 0.94%–3.32%, compared to the second-best method. The Luojia_VSRSet dataset facilitates further research in this domain. The proposed network generates high-quality, artifact-free high-resolution frames across all dynamic target scenes, enhancing the use of Luojia3-01 satellite video for related applications.

Hyperspectral remote sensing images offer a unique opportunity to quickly monitor water depth, but how to utilize the enriched spectral information and improve its spatial resolution remains a challenge. We proposed a water depth estimation framework to improve spatial resolution using deep learning and four inversion methods and verified the effectiveness of different super resolution and inversion methods in three waterbodies based on HJ-2 hyperspectral images. Results indicated that it was feasible to use HJ-2 hyperspectral images with a higher spatial resolution via super resolution methods to estimate water depth. Deep learning improves the spatial resolution of hyperspectral images from 48 m to 24 m and shows less information loss with peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and spectral angle mapper (SAM) values of approximately 37, 0.92, and 2.42, respectively. Among four inversion methods, the multilayer perceptron demonstrates superior performance for the water reservoir, achieving the mean absolute error (MAE) and the mean absolute percentage error (MAPE) of 1.292 m and 22.188%, respectively. For two rivers, the random forest model proves to be the best model, with an MAE of 0.750 m and an MAPE of 10.806%. The proposed method can be used for water depth estimation of different water bodies and can improve the spatial resolution of water depth mapping, providing refined technical support for water environment management and protection.

Reconstructing 3D spatial information from satellite images remains a pivotal challenge in photogrammetry. While traditional methods rely on stereo image pairs, recent advances in monocular height estimation (MHE) have demonstrated the promising feasibility of recovering 3D information without stereo observation conditions. At present, the mainstream deep learning-based MHE models typically learn the mapping between the image-derived deep features and height distribution from extensive RGB-height image pairs. However, the scarcity of satellite MHE datasets severely restricts the application of MHE models in real-world scenarios. In addition, existing MHE datasets predominantly rely on pairs of approximate orthophotography images, which can not be guaranteed in practical satellite imagery acquisitions, undoubtedly leading to the inability of the existing MHE models to learn elevation perception information from varying observation geometries. On the other hand, collecting multi-angle imagery data through cross-orbit or multi-temporal satellite observation introduces significant costs and temporal latency for producing such multi-angle datasets. To this end, this paper constructs the Multi-Angle Monocular Height Estimation Dataset (MAMHESet) to verify the performance of the MHE models when processing multi-angle images. The proposed dataset is constructed using multi-angle images acquired from the Luojia3-01 optical video satellite in gazing mode and the corresponding aerial LiDAR point clouds. We develop a novel dataset generation pipeline, which effectively mitigates dataset construction challenges caused by the inherent heterogeneous point cloud quality differences. To the best of our knowledge, this is the first MHE dataset derived from optical video satellite imagery with multiple observation angles. The extensive training and accuracy evaluations on several mainstream MHE models have demonstrated the practicality of this dataset, which further expands the application potential of the monocular satellite imagery in 3D reconstruction tasks.