Change detection (CD) is vital in remote sensing (RS) image interpretation applications and has been well improved driven by deep learning (DL). However, existing DL-based methods have poor model robustness due to the effect of limited number of change samples in the existing public datasets, and on the other hand, existing most CD methods only process the input image in the spatial domain, and insufficiently process it at the frequency domain level, which leads to its ability to obtain only shallow semantic information of the image, and insufficient ability to deal with the high-frequency information (edges, details) in the RS. In order to address the issue, this paper proposed DSFI-CD, which generates part of the pseudo-images by using conditional DDPM. Furthermore, it suggests a module called Spatial Frequency Interaction (SFI), which combines the extraction and fusion of the corresponding spatial and frequency domain data with the intersection of the dual time-phase features prior to and following the change to give the model more reliable information. Additionally, we designed the Edge Enhanced (EE) Module to further decompose the extracted features in the SFI Module into boundary and body information for the change region edge information. We do this by using the concept of flow field, and we also use the Laplace pyramid and multilevel feature fusion to increase the sensitivity of the model to the boundary while reducing the impact of noise in the non-change region. Comprehensive experiments on four publicly accessible CD datasets show that DSFI-CD performs better than other state-of-the-art (SOTA) methods.

One of the key objectives in tunnel illness detection is identifying tunnel lining leakage, and deep learning-based image semantic segmentation approaches can automatically locate tunnel lining leakage. However, in order to meet the real-time processing needs of professional mobile inspection equipment, existing leakage image segmentation approaches have difficulties in identifying real-time, dealing with voids, and dealing with edge discontinuities in the leaking zone. To address the aforementioned issues, this study introduces the PP-LiteSeg-Attn model, which takes the real-time semantic segmentation model PP-LiteSeg-B as baseline model, and combines the multi-layer CBAM attention mechanism and the CoT attention mechanism. Using the publically available dataset Water-Leakage, we trained and validated the PP-LiteSeg-Attn model, and attained IoU and F1 values of 88.18% and 93.72%, respectively, outperforming similar models in both measures. Extensive experiments show that the segmentation speed of the PP-LiteSeg-Attn model reaches 112.28 FPS, which meets real-time requirements, and that the model can effectively solve problems such as the appearance of voids in the seepage area, discontinuity, and fuzzy segmentation of seepage edges. The PP-LiteSeg-Attn model is better applicable to complicated tunnel settings, offering technical references for real-time diagnosis of tunnel illnesses.

The purpose of remote sensing image change detection (RSCD) is to detect differences between bitemporal images taken at the same place. Deep learning has been extensively used to RSCD tasks, yielding significant results in terms of result recognition. However, due to the shooting angle of the satellite, the impacts of thin clouds, and certain lighting conditions, the problem of fuzzy edges in the change region in some remote sensing photographs cannot be properly handled using current RSCD algorithms. To solve this issue, we proposed a body decouple multiscale by feature aggregation change detection , a novel model that collects both global and local feature map information in the channel and space dimensions of the feature map during the training and prediction phases. This approach allows us to successfully extract the change region's boundary information while also divorcing the change region's main body from its boundary. Numerous studies have shown that the assessment metrics and evaluation effects of the model described in this article on the publicly available datasets DSIFN-CD, S2Looking, and WHU-CD are the best when compared to other models.

Semantic change detection (SCD) in bi-temporal remote sensing images aims to localize and semantically classify changed regions, which is crucial for land cover monitoring and disaster assessment. Existing methods often struggle with spatial misalignment and semantic inconsistency between temporal domains. In this letter, we propose FAPMNet, a Flow-Aligned Prototype Memory Network, which integrates a Mamba-based Vision State Space Model (VSSM) encoder for efficient long-range spatial modeling. To address temporal misalignment, a Flow Alignment Module (FAM) estimates dense flow fields to warp and align features. A Prototype Memory Module (PMM) further enhances semantic consistency via class-guided feature reconstruction. Evaluated on the SECOND and Landsat-SCD datasets, FAPMNet achieves new state-of-the-art results with 74.48% mIoU and 66.04% F_scd on SECOND, and 90.04% mIoU and 90.20% F_scd on Landsat-SCD, surpassing recent Transformer and Mamba-based baselines.

Accurate building extraction (BE) from multimodal remote sensing imagery is critical to urban management and sustainable development. Although the integration of optical and synthetic aperture radar (SAR) data can provide complementary information, large-scale deployment is difficult due to blurring object boundaries, severe modality disparities, and expensive computing costs. To overcome these concerns, we proposed SSDFNet, a state space driven multimodal fusion framework for the Optical-SAR BE task. By involving VMamba into the backbone, SSDFNet captures long-term contextual dependencies with linear computing complexity, enabling efficient global modeling. To improve extraction quality, we proposed the Semantic Detail Fusion Module (SDFM), which injects high-level semantic cues into high-resolution features for finer structural delineation; the Dual Stream Fusion Module (DSFM), which adaptively aligns heterogeneous optical and SAR representations; and the Frequency Domain Enhancement Fusion Module (FDEFM), which improves edge sharpness and topological consistency through frequency decomposition and recombination. Extensive experiments on the DFC23 Track 2, MSAW, and INRIA datasets demonstrate the superiority of SSDFNet, achieving IoU scores of 85.31%, 80.70%, and 84.38%, respectively. In addition, we construct a 0.5 m resolution Optical-SAR rooftop dataset covering Singapore, on which SSDFNet attains an IoU of 80.41%. Furthermore, we quantitatively evaluate the effectiveness of the extracted building rooftops in a downstream solar photovoltaic (PV) potential estimation task by integrating rooftops with height information. The results demonstrate that improved geometric accuracy of building extraction directly leads to more reliable PV potential estimation, highlighting the practical significance of the proposed method for renewable energy planning.