ADMNet: adaptive deformable convolution large model combining multi-level progressive fusion for Building Change Detection

Geo-spatial Information Science, Journal Year: 2025, Volume and Issue: unknown, P. 1 - 18

Published: Jan. 17, 2025

Building Change Detection (BCD) based on high-resolution Remote Sensing Images (RSI) simplifies urban surface monitoring. Nevertheless, the mainstream detection methods utilizing traditional convolution and attention mechanisms are often prone to errors due loss of edge detail information underutilization global context information. To address these issues, this paper presents a large model, namely ADMNet, which is built adaptive deformable designed handles various types building change First, we propose Siamese neural network (ADC) modules. The ADC module incorporates spatial offset parameters into convolutional kernel sampling mapping weights capture irregularly varying features for local receptive fields. Second, utilize model semantically driven enhance awareness construct long-range feature dependencies from multi-scale information, then integrated with locally structure achieve accurate localization. Furthermore, design Multi-Level Progressive Feature Fusion (MLPFF) that enhances characterization capabilities ensure internal integrity improves performance by integrating priori knowledge large-model transfer learning. evaluate effectiveness generalizability conduct comparative experiments current two datasets, LEVIR-CD WHU-CD, land cover dataset, SYSU-CD. results show ADMNet outperforms all methods. source code available at https://github.com/spaceYu180/ADMNet.

Language: Английский

---

DAM-Net: Flood detection from SAR imagery using differential attention metric-based vision transformers

ISPRS Journal of Photogrammetry and Remote Sensing, Journal Year: 2024, Volume and Issue: 212, P. 440 - 453

Published: May 20, 2024

Language: Английский

---

SCD-SAM: Adapting Segment Anything Model for Semantic Change Detection in Remote Sensing Imagery

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2024, Volume and Issue: 62, P. 1 - 13

Published: Jan. 1, 2024

Semantic change detection (SCD) has gradually emerged as a prominent research focus in remote sensing image processing due to its critical role earth observation applications. In view of powerful semantic-driven feature extraction capability, the Segment Anything Model (SAM) demonstrated suitability across various visual scenes. However, it suffers from significant performance degradation when confronted with images, especially those containing ground objects that possess inter-class similarity and substantial intra-class variations. To address above issues, we propose SCD-SAM, aiming leverage potent recognition capabilities SAM for enhanced accuracy robustness SCD. Specifically, introduce contextual semantic change-aware dual encoder combines MobileSAM CNN extract progressive features parallel, inject local into through depth interaction compensate Transformer's limitations perceiving details. Besides, order utilize strong capability adaptor aggregates semantic-oriented information about changing objects. better integrate extracted information, devise aggregation decoder binary respectively, alleviating gap different scales. The quantitative results show SCD-SAM outperforms state-of-the-art SCD methods on publicly open datasets (e.g., SECOND-CD Landsat-CD). code will be made available at https://github.com/yzygit1230/SCD-SAM.

Language: Английский

---

GS-DeepLabV3+: A Mountain Tea Disease Segmentation Network Based on Improved Shuffle Attention and Gated Multidimensional Feature Extraction

Crop Protection, Journal Year: 2024, Volume and Issue: 183, P. 106762 - 106762

Published: May 25, 2024

Language: Английский

---

High-precision flood detection and mapping via multi-temporal SAR change analysis with semantic token-based transformer

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2024, Volume and Issue: 131, P. 103991 - 103991

Published: June 24, 2024

Language: Английский

---

Rethinking Building Change Detection: Dual-Frequency Learnable Visual Encoder With Multiscale Integration Network

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Journal Year: 2024, Volume and Issue: 17, P. 6174 - 6188

Published: Jan. 1, 2024

Remote sensing (RS) image change detection methods based on deep learning such as convolutional neural networks (CNN) and transformers are still spatial domain-based processing by nature, their accuracy is strongly affected chromatic aberration due to imaging time, shadows caused lighting conditions, object confusion other disturbances. In this study, we revisit (CD) from a signal perspective, framing it the task of consistency distributional features two 2D signals. We aim extract primary components signals while suppressing interfering noises. To address this, propose novel CD method called DFNet, which leverages dual-frequency learnable encoder. First, construct feature encoder Siamese framework capture local high-frequency global low-frequency using CNN attention mechanisms after dividing input RS into channels. Second, introduce frequency explicit visual center (FEVC) module part multifrequency domain dense interaction (MFDDI) at decoder stage, allowing long-distance dependency be established between high-low in same layer well aggregation regions small edge variations. addition, MFDDI adopts layer-by-layer interactive fusion approach synthesize discriminative information wide range, enhancing characterization capability conduct comparison experiments with current mainstream land cover dataset SYSU-CD building datasets, LEVIR-CD WHU-CD, results show that our not only resistant interference but also outperforms all methods.

Language: Английский

---

InCR: Inception and concatenation residual block-based deep learning network for damaged building detection using remote sensing images

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2023, Volume and Issue: 123, P. 103483 - 103483

Published: Sept. 1, 2023

In February 2023, Turkey experienced a series of earthquakes that caused significant damage to buildings and affected many people. Detecting building quickly is crucial for helping earthquake victims, we believe machine learning models offer promising solution. our research, introduce new, lightweight deep-learning model capable accurately classifying damaged in remote-sensing datasets. Our main goal create an automated detection system using novel model. We started by collecting new dataset with two categories: undamaged buildings. Then, developed unique convolutional neural network (CNN) called the inception concatenation residual (InCR) deep network, which incorporates concatenation-based blocks improve performance. trained InCR on newly collected used it extract features from images global average pooling. To refine these select most informative ones, applied iterative neighborhood component analysis (INCA). Finally, classified refined commonly shallow classifiers. evaluate method, tenfold cross-validation (10-fold CV) eight The results showed all classifiers achieved classification accuracies higher than 98 %. This demonstrates proposed viable option CNNs can be accurate application. research presents solution challenge after earthquakes, showing highlight potential approach.

Language: Английский

---

Multistage Interaction Network for Remote Sensing Change Detection

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(6), P. 1077 - 1077

Published: March 19, 2024

Change detection in remote sensing imagery is vital for Earth monitoring but faces challenges such as background complexity and pseudo-changes. Effective interaction between bitemporal images crucial accurate change information extraction. This paper presents a multistage network designed effective detection, incorporating at the image, feature, decision levels. At image level, directly extracted from intensity changes, mitigating potential loss during feature Instead of separately extracting features images, feature-level jointly extracts images. By enhancing relevance to spatial variant shared semantic channels, excels overcoming The decision-level combines image-level interactions, producing multiscale differences precise prediction. Extensive experiments demonstrate superior performance our method compared existing approaches, establishing it robust solution detection.

Language: Английский

---

Cross-Visual Style Change Detection for Remote Sensing Images via Representation Consistency Deep Supervised Learning

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(5), P. 798 - 798

Published: Feb. 25, 2025

Change detection techniques, which extract different regions of interest from bi-temporal remote sensing images, play a crucial role in various fields such as environmental protection, damage assessment, and urban planning. However, visual style interferences stemming varying acquisition times, radiation, weather, phenology changes, often lead to false detections. Existing methods struggle robustly measure background similarity the presence discrepancies lack quantitative validation for assessing their effectiveness. To address these limitations, we propose Representation Consistency Detection (RCCD), novel deep learning framework that enforces global local spatial consistency features across encoding decoding stages robust cross-visual change detection. RCCD leverages large-kernel convolutional supervision context awareness content-aware transfer feature harmonization, effectively suppressing interference variations. Extensive evaluations on S2Looking LEVIR-CD+ datasets demonstrate RCCD’s superior performance, achieving state-of-the-art F1-scores. Furthermore, dedicated subsets with large differences, exhibits more substantial improvements, highlighting its effectiveness mitigating caused by errors. The code has been open-sourced GitHub.

Language: Английский

---

Quantifying ecosystem service loss from wind farm-induced landscape fragmentation using remote sensing

Geocarto International, Journal Year: 2025, Volume and Issue: 40(1)

Published: May 29, 2025

Language: Английский

---