7 February 2025 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
In medium-resolution remote sensing images, offshore platforms are easily confused with other offshore objects, such as ships and offshore wind power equipment. In this study, an object detection network named OPDNet, which is based on a two-branch pseudo-Siamese structure, is proposed to accurately position and extract offshore platforms. OPDNet utilizes paired yet unsynchronized optical and radar images derived from the Sentinel 1 and 2 satellites as its inputs, leveraging the complementary nature of bitemporal and multimodal image data to compensate for the limitations inherent in single-modal images. On this basis, the feature extraction and classification capabilities of the network are improved for small objects by fusing a C2f feature pyramid integrated with the CloAttention module (C2f_CloAtt), context attention module (CAM), and space-depth convolution (SPD_Conv) modules. Experiments show that OPDNet yields significant object detection accuracy improvements over the comparative unimodal single-phase and unimodal multiphase methods, with an average precision (AP) value of 94.60% and an F1 score of 87.83%. Thus, OPDNet provides a powerful solution for large-scale remote sensing-based offshore platform monitoring.
17 May 2024 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Recent building-oriented change detection studies considered only morphological changes in buildings, and few publicly available change detection datasets further distinguish among building types. In this study, we propose a semantic change detection network named BSCDNet that considers both morphological and semantic changes in buildings. BSCDNet adopts a multitasking branch structure with object classification, change detection and segmentation to simultaneously achieve object-level semantic classification and change analysis of buildings in bitemporal, high-spatial-resolution (HSR) imagery. In the object classification branch, a graph attention network (GAT) is utilized to capture the spatial and semantic correlations among buildings during classification. The change detection branch applies both spatial and channel attention mechanisms to eliminate nonbuilding interference and enhance change features. Moreover, the segmentation branch adopts a distinctive instance segmentation procedure that improves the accuracy of object segmentation. We created a building change detection dataset with category differentiation based on HSR imagery to validate the proposed method. Ablation experiments verify the effectiveness and advantages of the above task branches. Furthermore, in comparative experiments with several SOTA semantic change detection methods, such as HRSCD, SCDNet and MR_CD, BSCDNet reached the optimal level in terms of F1 and mIoU when evaluating the change detection performance, as well as kappa and score for evaluating the classification performance.
12 May 2026 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Raft aquaculture facilities refer to cultivation structures composed of floats and raft ropes, which form floating raft platforms for the suspended culture of economically valuable seaweeds, sessile animals, and other aquatic species. Typically, multiple raft facilities are systematically arranged to establish raft aquaculture areas (RAAs). Accurately mapping the spatial distribution of RAAs is critical for marine ranching planning, yield prediction, and ecological environmental assessment. Most existing studies utilize medium- to high-resolution remote sensing imagery to extract RAAs on the basis of spectral differences between rafts and surrounding seawater. However, factors such as varying growth stages and partial harvesting often result in weak spectral signals for some RAAs, making them difficult to distinguish from adjacent water bodies and leading to suboptimal extraction accuracy. To address these challenges, we propose SERaftFormer, a novel deep learning-based method that leverages submeter very high-resolution (VHR) imagery. SERaftFormer contributes to improved extraction performance for weak-featured raft aquaculture areas (WFRAAs) by explicitly integrating the edge features of internal raft strips into a region-based semantic segmentation framework. Specifically, the proposed wavelet-based transformer block with edge attention (WTBEA) is designed to simultaneously enhance edge features and effectively suppress background noise. The raft feature attention module (RFAM) refines the raft features by integrating internal strip edge features through the reverse attention mechanism. Additionally, the loss function incorporates an edge loss term, which provides supplementary supervision focused on edge features. The experimental results demonstrate that SERaftFormer improves the extraction performance of WFRAAs, mitigates adhesion artifacts, and outperforms edge-aware models and specialized raft extraction models (e.g., SEANet and RADNet) in method accuracy. The relevant code and dataset will be publicly released on https://github.com/ShikangTao/SERaftFormer.