Sea ice extraction and segmentation of remote sensing images is the basis for sea ice monitoring. Traditional image segmentation methods rely on manual sampling and require complex feature extraction. Deep-learning-based semantic segmentation methods have the advantages of high efficiency, intelligence, and automation. Sea ice segmentation using deep learning methods faces the following problems: in terms of datasets, the high cost of sea ice image label production leads to fewer datasets for sea ice segmentation; in terms of image quality, remote sensing image noise and severe weather conditions affect image quality, which affects the accuracy of sea ice extraction. To address the quantity and quality of the dataset, this study used multiple data augmentation methods for data expansion. To improve the semantic segmentation accuracy, the SC-U²-Net network was constructed using multiscale inflation convolution and a multilayer convolutional block attention module (CBAM) attention mechanism for the U²-Net network. The experiments showed that (1) data augmentation solved the problem of an insufficient number of training samples to a certain extent and improved the accuracy of image segmentation; (2) this study designed a multilevel Gaussian noise data augmentation scheme to improve the network’s ability to resist noise interference and achieve a more accurate segmentation of images with different degrees of noise pollution; (3) the inclusion of a multiscale inflation perceptron and multilayer CBAM attention mechanism improved the ability of U²-Net network feature extraction and enhanced the model accuracy and generalization ability.

For addressing the technical challenges of photon denoising and water level extraction in ICESat-2 satellite-based water monitoring applications, this paper proposes an innovative solution integrating Gaussian function fitting with Hough transform. The method first employs histogram Gaussian fitting to achieve coarse denoising of water body regions. Subsequently, a probability attenuation model based on elevation differences between adjacent photons is constructed to accomplish refined denoising through iterative optimization of adaptive thresholds. Building upon this foundation, the Hough transform technique from image processing is introduced into photon cloud processing, enabling robust water level extraction from ICESat-2 data. Through rasterization, discrete photon distributions are converted into image space, where straight lines conforming to the photon distribution are then mapped as intersection points of sinusoidal curves in Hough space. Leveraging the noise-resistant characteristics of the Hough space accumulator, the interference from residual noise photons is effectively eliminated, thereby achieving high-precision water level line extraction. Experiments were conducted across five typical water bodies (Qinghai Lake, Long Land, Ganquan Island, Qilian Yu Islands, and Miyun Reservoir). The results demonstrate that the proposed denoising method outperforms DBSCAN and OPTICS algorithms in terms of accuracy, precision, recall, F1-score, and computational efficiency. In water level estimation, the absolute error of the Hough transform-based line detection method remains below 2 cm, significantly surpassing the performance of mean value, median value, and RANSAC algorithms. This study provides a novel technical framework for effective global water level monitoring.

The development of satellite remote sensing technology has made it easier to obtain satellite imagery. Compared to imagery obtained from aerial photography, satellite imagery has the advantages of wide coverage, high acquisition efficiency, and periodic revisits. In photogrammetry, the 3D reconstruction technology of satellite imagery often requires optimization and adjustment of numerous RPC parameters, which to some extent limits the speed and accuracy of 3D reconstruction. At the same time, the progress in 3D reconstruction technology in the field of computer vision has shown certain advantages in terms of accuracy and speed. However, these methods are specifically designed for pinhole imaging models and cannot be directly applied to the 3D reconstruction of satellite imagery with row-sampled central projection. The introduction of the equivalent pinhole imaging model enables computer vision methods to perform 3D reconstruction on satellite imagery. This local approximation introduces re-projection errors when the RPC model is equivalent to the pinhole imaging model, thereby affecting the accuracy of 3D reconstruction.This paper investigates the causes and patterns of re-projection errors in the equivalent pinhole imaging model and proposes a method for generating pseudo-images through iterative resampling, as well as a method for partitioning satellite images to equivalently approximate the pinhole imaging model. Test results on the MVS3D dataset show that both methods can reduce re-projection errors, thereby improving the accuracy of 3D reconstruction of satellite images using the equivalent pinhole imaging model.

To address the challenges of multi-scale object recognition and complex background in-terference in remote sensing images, this paper proposes a lightweight improved algo-rithm based on YOLOv8-obb. By integrating a hybrid local-channel attention mechanism (MLCA), dynamic upsampling (DySample), and a reparameterized cross-scale feature aggregation module (RepNCSPELAN), the algorithm achieves dual enhancements in de-tection accuracy and computational efficiency. The MLCA module enhances feature dis-criminability in complex backgrounds through a dual-path mechanism combining local and global pooling. The DySample module employs content-aware sampling point gen-eration to improve feature reconstruction for multi-scale targets. The RepNCSPELAN module reduces model parameters by 29% while preserving cross-scale feature fusion ca-pabilities. Experimental results on the DOTA dataset demonstrate a 2.3% improvement in mAP50, with parameters reduced to 2.2M and FLOPs decreased by 27%. Cross-dataset validation on DIOR further confirms a 1.5% mAP50 gain. Compared to mainstream lightweight models (e.g., YOLOv5n, YOLOv8n), the proposed algorithm exhibits superior performance in accuracy (65.3% vs. 60.8%), parameter efficiency (2.2M vs. 3.1M), and computational cost (6.2G vs. 8.5G FLOPs). Ablation studies validate the efficacy of each module, while visualizations highlight robustness in dense small-object detection and ro-tated target localization. This work provides an efficient solution for real-time remote sensing object detection in complex scenarios and offers a novel technical pathway for lightweight deep learning model design.

Platform jitter in stereo mapping satellites introduces periodic stripe artifacts into digital surface models (DSMs), degrading geometric quality, while existing detection methods usually depend on disparity maps or high-frequency attitude data. This study proposes a DSM-based frequency-domain framework for detecting and removing jitter-induced stripes in GF-7 DSMs. First, 2D Fourier narrow-band notch analysis of detrended high-pass DSMs estimates stripe orientation and dominant period from directional and radial spectral peak prominence, and a Gaussian narrow-band stop filter is constructed for global suppression. Then, under these spectral priors, a profile-template method uses 1D median profiles, IIR notch filtering, and a global stripe template with a slowly varying amplitude field to model and remove the stripe component in the spatial domain. Multi-temporal co-registered DSM differencing provides reference stripe characteristics for evaluation. Experiments on three GF-7 DSMs show a stripe normal direction aligned with the subsatellite ground track, a stable dominant stripe period of about 90 m, period errors not exceeding 0.34 m, and consistency above 99.7% between the two methods. The spectral notch filtering suppresses more than 92% of stripe-band energy while preserving the overall spectral shape. The proposed framework therefore enables accurate jitter characterization and effective destriping without attitude or disparity information.