Unsupervised Change Detection Research Articles

Unsupervised Change Detection Methods Applied to Landslide Mapping: Case Study in São Sebastião, Brazil

ABSTRACTLandslides represent a growing global geological hazard, further intensified by climate‐induced changes. Remote sensing data, through its capacity for repetitive collection and change detection techniques, that compare and quantify the spatio‐temporal alterations over time, plays a critical role in landslide detection. Considering the February 2023 São Sebastião event and Sentinel‐2 imagery, we assessed diverse unsupervised change detection techniques, encompassing both traditional and recent machine learning‐based approaches. Notably, the Floating References (FR) and Homogeneous Blocks Single‐class Classification (HBSC) methods outperform classic approaches and deliver the most accurate results with F1‐Score and kappa coefficient exceeding 0.96 and 0.92, respectively. These outcomes demonstrate the efficacy of machine learning in automating landslide delineation and underscore the necessity of meticulous data and parameter selection in achieving high‐accuracy automatic landslide mapping. Lastly, this study fills a significant gap in the existing literature by evaluating unsupervised change detection methods for landslide mapping within the Brazilian context.

Unsupervised change detection for remotely sensed multi-spectral images based on context-aware saliency-spectral-spatial features and weighted coarse-to-fine fusion

Unsupervised change detection for remotely sensed multi-spectral images based on context-aware saliency-spectral-spatial features and weighted coarse-to-fine fusion

Unsupervised change detection based on image reconstruction loss with segment anything

ABSTRACT In remote sensing, change detection based on deep learning shows promising performance. However, collecting multi-temporal paired images for training a change detection model is costly. To solve this problem, unsupervised change detection methodologies have been proposed, but their performance is still low. In this article, we introduce the Unsupervised Change Detection Based on Image Reconstruction Loss (CDRL) and CDRL with Segment Anything (CDRL-SA). Requiring only single-temporal unlabelled images for training, CDRL uses the original and a photometrically transformed image as an unchanged pair, input to a bi-temporal transformer-based network for reconstructing the original image. During inference, changed pairs result in significant reconstruction loss, highlighting change areas. To capture finer details, we change the structure of CDRL to a transformer-based model and introduce the CutSwap method for effective training. Furthermore, this output is fused with the results of a recently proposed Segment Anything (SA) model to improve the final output. We assessed the performance of CDRL and CDRL-SA using the LEVIR change detection dataset and CLCD dataset, and the method achieved competitive results of 88.9 and 91.6 ACC for the respective datasets, demonstrating its effectiveness in unsupervised change detection tasks.

Mapping fine-scale seagrass disturbance using bi-temporal UAV-acquired images and multivariate alteration detection

Seagrasses provide critical ecosystem services but cumulative human pressure on coastal environments has seen a global decline in their health and extent. Key processes of anthropogenic disturbance can operate at local spatio-temporal scales that are not captured by conventional satellite imaging. Seagrass management strategies to prevent longer-term loss and ensure successful restoration require effective methods for monitoring these fine-scale changes. Current seagrass monitoring methods involve resource-intensive fieldwork or recurrent image classification. This study presents an alternative method using iteratively reweighted multivariate alteration detection (IR-MAD), an unsupervised change detection technique originally developed for satellite images. We investigate the application of IR-MAD to image data acquired using an unoccupied aerial vehicle (UAV). UAV images were captured at a 14-week interval over two seagrass beds in Brisbane Water, NSW, Australia using a 10-band Micasense RedEdge-MX Dual camera system. To guide sensor selection, a further three band subsets representing simpler sensor configurations (6, 5 and 3 bands) were also analysed using eight categories of seagrass change. The ability of the IR-MAD method, and for the four different sensor configurations, to distinguish the categories of change were compared using the Jeffreys-Matusita (JM) distance measure of spectral separability. IR-MAD based on the full 10-band sensor images produced the highest separability values indicating that human disturbances (propeller scars and other seagrass damage) were distinguishable from all other change categories. IR-MAD results for the 6-band and 5-band sensors also distinguished key seagrass change features. The IR-MAD results for the simplest 3-band sensor (an RGB camera) detected change features, but change categories were not strongly separable from each other. Analysis of IR-MAD weights indicated that additional visible bands, including a coastal blue band and a second red band, improve change detection. IR-MAD is an effective method for seagrass monitoring, and this study demonstrates the potential for multispectral sensors with additional visible bands to improve seagrass change detection.

Large-scale flood modeling and forecasting with FloodCast

Large-scale hydrodynamic models generally rely on fixed-resolution spatial grids and model parameters as well as incurring a high computational cost. This limits their ability to accurately forecast flood crests and issue time-critical hazard warnings. In this work, we build a fast, stable, accurate, resolution-invariant, and geometry-adaptive flood modeling and forecasting framework that can perform at large scales, namely FloodCast. The framework comprises two main modules: multi-satellite observation and hydrodynamic modeling. In the multi-satellite observation module, a real-time unsupervised change detection method and a rainfall processing and analysis tool are proposed to harness the full potential of multi-satellite observations in large-scale flood prediction. In the hydrodynamic modeling module, a geometry-adaptive physics-informed neural solver (GeoPINS) is introduced, benefiting from the absence of a requirement for training data in physics-informed neural networks (PINNs) and featuring a fast, accurate, and resolution-invariant architecture with Fourier neural operators. To adapt to complex river geometries, we reformulate PINNs in a geometry-adaptive space. GeoPINS demonstrates impressive performance on popular partial differential equations across regular and irregular domains. Building upon GeoPINS, we propose a sequence-to-sequence GeoPINS model to handle long-term temporal series and extensive spatial domains in large-scale flood modeling. This model employs sequence-to-sequence learning and hard-encoding of boundary conditions. Next, we establish a benchmark dataset in the 2022 Pakistan flood using a widely accepted finite difference numerical solution to assess various flood simulation methods. Finally, we validate the model in three dimensions - flood inundation range, depth, and transferability of spatiotemporal downscaling - utilizing SAR-based flood data, traditional hydrodynamic benchmarks, and concurrent optical remote sensing images. Traditional hydrodynamics and sequence-to-sequence GeoPINS exhibit exceptional agreement during high water levels, while comparative assessments with SAR-based flood depth data show that sequence-to-sequence GeoPINS outperforms traditional hydrodynamics, with smaller simulation errors. The experimental results for the 2022 Pakistan flood demonstrate that the proposed method enables high-precision, large-scale flood modeling with an average MAPE of 14.93 % and an average Mean Absolute Error (MAE) of 0.0610 m for 14-day water depth simulations while facilitating real-time flood hazard forecasting using reliable precipitation data.

Heuristic approaches for non-exhaustive pattern-based change detection in dynamic networks

AbstractDynamic networks are ubiquitous in many domains for modelling evolving graph-structured data and detecting changes allows us to understand the dynamic of the domain represented. A category of computational solutions is represented by the pattern-based change detectors (PBCDs), which are non-parametric unsupervised change detection methods based on observed changes in sets of frequent patterns over time. Patterns have the ability to depict the structural information of the sub-graphs, becoming a useful tool in the interpretation of the changes. Existing PBCDs often rely on exhaustive mining, which corresponds to the worst-case exponential time complexity, making this category of algorithms inefficient in practice. In fact, in such a case, the pattern mining process is even more time-consuming and inefficient due to the combinatorial explosion of the sub-graph pattern space caused by the inherent complexity of the graph structure. Non-exhaustive search strategies can represent a possible approach to this problem, also because not all the possible frequent patterns contribute to changes in the time-evolving data. In this paper, we investigate the viability of different heuristic approaches which prevent the complete exploration of the search space, by returning a concise set of sub-graph patterns (compared to the exhaustive case). The heuristics differ on the criterion used to select representative patterns. The results obtained on real-world and synthetic dynamic networks show that these solutions are effective, when mining patterns, and even more accurate when detecting changes.

Self-Guided Autoencoders for Unsupervised Change Detection in Heterogeneous Remote Sensing Images

Self-Guided Autoencoders for Unsupervised Change Detection in Heterogeneous Remote Sensing Images

Detecting and Monitoring Artisanal Mining Operations in Semi-Arid Terrain Using Multitemporal SAR Data for InSAR Coherence Estimation and Unsupervised Classification

Abstract. The impact of artisanal mining in sub-Saharan Africa and other developing countries are often huge as they leave trails of environmental and socio-economic effects on the people and the environment. 123 Multi-temporal SAR data acquired between 2019 and 2023 was used. SARPROZ was used to co-register the images and other preprocessing steps like radiometric and geometric correction to transform the images from SAR coordinates to geographic coordinates after generating all the interferograms. A stack of the 122 coherence map was created. Unsupervised classification was implemented on the stack. Principal component analysis of dimensionality reductions method yielded a far better result than the other unsupervised cluster methods attempted, it showed a very high classification accuracy of the terrain. The principal component analysis worked by computing the covariance matrix of the stacked coherence map then performed eigen-decomposition on it to yield eigenvectors and eigenvalues, The eigenvectors corresponding to the largest eigenvalues represent the principal components. These principal components capture the directions of maximum variance, and the eigenvectors provides a reduced-dimensional representation of the image stack which is then used to reconstruct an approximation of the original image that captures the essential features of the original SAR data. Backscatter intensity of the SAR images processed for this study period for unsupervised change detection and land cover classification, delineated the different features and classes based on long term coherence values.

Is the change deforestation? Using time-series analysis of satellite data to disentangle deforestation from other forest degradation causes

Protecting natural ecosystems requires monitoring approaches that work as early warning systems to avoid degradation and protect biodiversity. However, separating forest disturbance causes in change-detection pipelines is challenging due to the complex interplay of multiple drivers affecting vegetation. This study aims to detect deforestation in highly heterogeneous ecosystems. We used Landsat NDVI time-series data for testing three unsupervised change detection methods: (1) the non-parametric phenological anomaly detection (npphen), (2) the continuous change detection and classification (CCDC), and (3) the pruned exact linear time (PELT) algorithms. We used visual interpretation of Google Earth Pro high-resolution data (<10 m) to depict deforestation, and natural-induced changes, like forest browning and fires, evaluating the performance of the unsupervised methods. Additionally, a Random Forest model trained with the outputs from detection algorithms together with elevation and radar vegetation index data were utilised to depict deforestation in a second step. While PELT slightly outperformed other methods for tracking general vegetation changes, with overall accuracies (OA) ranging from 0.78 to 0.99, depending on the vegetation type, it also showed the slowest deforestation tracking response. CCDC presented the fastest response and an OA between 0.78 and 0.95. Additionally, we observed a mean OA of only 0.47 when separating deforestation from other changes using only the unsupervised models. On the other hand, deforestation was accurately detected (OA = 0.93; kappa = 0.83) when using CCDC outputs within a secondary supervised classification, agreeing with selected citizen-based complaints from the Environmental Superintendence. The relatively fast response in deforestation tracking using CCDC makes it a viable alternative for near real-time monitoring. Commonly used unsupervised detection methods may be coupled with supervised techniques to depict vegetation change sources robustly. This application constitutes a step forward for managing and monitoring vegetation areas in highly complex and dynamic landscapes, like Mediterranean ecosystems.

Unsupervised Change Detection in HR Remote Sensing Imagery Based on Local Histogram Similarity and Progressive Otsu

Unsupervised change detection of land cover in multispectral satellite remote sensing images with a spatial resolution of 2–5 m has always been a challenging task. This paper presents a method of detecting land cover changes in high-spatial-resolution remote sensing imagery. This method has three characteristics: (1) Extended center-symmetric local binary pattern (XCS-LBP) is used to extract image features to emphasize spatial context information in initial change detection. Then, spectral information is combined to improve the accuracy of change detection. (2) The local histogram distance of XCS-LBP features is used as the change vector to improve the expression of change information. (3) A progressive Otsu method is developed for threshold segmentation of the change vector to reduce the false detection rate. Four datasets with different landscape complexities and seven state-of-the-art unsupervised change detection methods were used to test the performance of the proposed method. Quantitative results showed that the proposed method reduced the false detection rate and improved the accuracy of the detection of land cover changes. The F1 score achieved by the proposed method reached 0.8688, 0.8867, 0.7725, and 0.6634, respectively, which are higher than the highest corresponding F1 score achieved by the benchmark methods (0.8533, 0.8549, 0.6545, and 0.5895, respectively).

LIRRN: Location-Independent Relative Radiometric Normalization of Bitemporal Remote-Sensing Images.

Relative radiometric normalization (RRN) is a critical pre-processing step that enables accurate comparisons of multitemporal remote-sensing (RS) images through unsupervised change detection. Although existing RRN methods generally have promising results in most cases, their effectiveness depends on specific conditions, especially in scenarios with land cover/land use (LULC) in image pairs in different locations. These methods often overlook these complexities, potentially introducing biases to RRN results, mainly because of the use of spatially aligned pseudo-invariant features (PIFs) for modeling. To address this, we introduce a location-independent RRN (LIRRN) method in this study that can automatically identify non-spatially matched PIFs based on brightness characteristics. Additionally, as a fast and coregistration-free model, LIRRN complements keypoint-based RRN for more accurate results in applications where coregistration is crucial. The LIRRN process starts with segmenting reference and subject images into dark, gray, and bright zones using the multi-Otsu threshold technique. PIFs are then efficiently extracted from each zone using nearest-distance-based image content matching without any spatial constraints. These PIFs construct a linear model during subject-image calibration on a band-by-band basis. The performance evaluation involved tests on five registered/unregistered bitemporal satellite images, comparing results from three conventional methods: histogram matching (HM), blockwise KAZE, and keypoint-based RRN algorithms. Experimental results consistently demonstrated LIRRN's superior performance, particularly in handling unregistered datasets. LIRRN also exhibited faster execution times than blockwise KAZE and keypoint-based approaches while yielding results comparable to those of HM in estimating normalization coefficients. Combining LIRRN and keypoint-based RRN models resulted in even more accurate and reliable results, albeit with a slight lengthening of the computational time. To investigate and further develop LIRRN, its code, and some sample datasets are available at link in Data Availability Statement.

CD-GAN: A robust fusion-based generative adversarial network for unsupervised remote sensing change detection with heterogeneous sensors

In the context of Earth observation, change detection boils down to comparing images acquired at different times by sensors of possibly different spatial and/or spectral resolutions or different modalities (e.g., optical or radar). Even when considering only optical images, this task has proven to be challenging as soon as the sensors differ by their spatial and/or spectral resolutions. This paper proposes a novel unsupervised change detection method dedicated to images acquired by such so-called heterogeneous optical sensors. It capitalizes on recent advances which formulate the change detection task into a robust fusion framework. Adopting this formulation, the work reported in this paper shows that any off-the-shelf network trained beforehand to fuse optical images of different spatial and/or spectral resolutions can be easily complemented with a network of the same architecture and embedded into an adversarial framework to perform change detection. A comparison with state-of-the-art change detection methods demonstrates the versatility and the effectiveness of the proposed approach.

Edge-Bound Change Detection in Multisource Remote Sensing Images

Detecting changes in multisource heterogeneous images is a great challenge for unsupervised change detection methods. Image-translation-based methods, which transform two images to be homogeneous for comparison, have become a mainstream approach. However, most of them primarily rely on information from unchanged regions, resulting in networks that cannot fully capture the connection between two heterogeneous representations. Moreover, the lack of a priori information and sufficient training data makes the training vulnerable to the interference of changed pixels. In this paper, we propose an edge-oriented generative adversarial network (EO-GAN) for change detection that indirectly translates images using edge information, which serves as a core and stable link between heterogeneous representations. The EO-GAN is composed of an edge extraction network and a reconstructive network. During the training process, we ensure that the edges extracted from heterogeneous images are as similar as possible through supplemented data based on superpixel segmentation. Experimental results on both heterogeneous and homogeneous datasets demonstrate the effectiveness of our proposed method.

Enhanced Graph Structure Representation for Unsupervised Heterogeneous Change Detection

Heterogeneous change detection (CD) is widely applied in various fields such as urban planning, environmental monitoring, and disaster management. It enhances the accuracy and comprehensiveness of surface change monitoring by integrating multi-sensor remote sensing data. Scholars have proposed many graph-based methods to address the issue of incomparable heterogeneous images caused by imaging differences. However, these methods often overlook the influence of changes in vertex status on the graph structure, which limits their ability to represent image structural features. To tackle this problem, this paper presents an unsupervised heterogeneous CD method based on enhanced graph structure representation (EGSR). This method enhances the representation capacity of the graph structure for image structural features by measuring the unchanged probabilities of vertices, thereby making it easier to detect changes in heterogeneous images. Firstly, we construct the graph structure using image superpixels and measure the structural graph differences of heterogeneous images in the same image domain. Then, we calculate the unchanged probability of each vertex in the structural graph and reconstruct the graph structure using this probability. To accurately represent the graph structure, we adopt an iterative framework for enhancing the representation of the graph structure. Finally, at the end of the iteration, the final change map (CM) is obtained by binary segmentation of the graph vertices based on their unchanged probabilities. The effectiveness of this method is validated through experiments on four sets of heterogeneous image datasets and two sets of homogeneous image datasets.

COMIC: An unsupervised change detection method for heterogeneous remote sensing images based on copula mixtures and Cycle-Consistent Adversarial Networks

In this paper, we consider the problem of change detection (CD) with two heterogeneous remote sensing (RS) images. For this problem, an unsupervised change detection method has been proposed recently based on the image translation technique of Cycle-Consistent Adversarial Networks (CycleGANs), where one image is translated from its original modality to the modality of the other image so that the difference map can be obtained by performing arithmetical subtraction. However, the difference map derived from subtraction is susceptible to image translation errors, in which case the changed area and the unchanged area are less distinguishable. To overcome the above shortcoming, we propose a new unsupervised copula mixture and CycleGAN-based CD method (COMIC), which combines the advantages of copula mixtures on statistical modeling and the advantages of CycleGANs on data mining. In COMIC, the pre-event image is first translated from its original modality to the post-event image modality. After that, by constructing a copula mixture, the joint distribution of the features from the heterogeneous images can be learnt according to quantitive analysis of the dependence structure based on the translated image and the original pre-event image, which are of the same modality and contain totally the same objects. Then, we model the CD problem as a binary hypothesis testing problem and derive its test statistics based on the constructed copula mixture. Finally, the difference map can be obtained from the test statistics and the binary change map (BCM) is generated by K-means clustering. We perform experiments on real RS datasets, which demonstrate the superiority of COMIC over the state-of-the-art methods.

Change Representation and Extraction in Stripes: Rethinking Unsupervised Hyperspectral Image Change Detection With an Untrained Network.

Deep learning-based hyperspectral image (HSI) change detection (CD) approaches have a strong ability to leverage spectral-spatial-temporal information through automatic feature extraction, and currently dominate in the research field. However, their efficiency and universality are limited by the dependency on labeled data. Although the newly applied untrained networks can avoid the need for labeled data, their feature volatility from the simple difference space easily leads to inaccurate CD results. Inspired by the interesting finding that salient changes appear as bright "stripes" in a new feature space, we propose a novel unsupervised CD method that represents and models changes in stripes for HSIs (named as StripeCD), which integrates optimization modeling into an untrained network. The StripeCD method constructs a new feature space that represents change features in stripes and models them in a novel optimization manner. It consists of three main parts: 1) dual-branch untrained convolutional network, which is utilized to extract deep difference features from bitemporal HSIs and combined with a two-stage channel selection strategy to emphasize the important channels that contribute to CD. 2) multiscale forward-backward segmentation framework, which is proposed for salient change representation. It transforms deep difference features into a new feature space by exploiting the structure information of ground objects and associates salient changes with the stripe-shaped change component. 3) stripe-shaped change extraction model, which characterizes the global sparsity and local discontinuity of salient changes. It explores the intrinsic properties of deep difference features and constructs model-based constraints to better identify changed regions in a controllable manner. The proposed StripeCD method outperformed the state-of-the-art unsupervised CD approaches on three widely used datasets. In addition, the proposed StripeCD method indicates the potential for further investigation of untrained networks in facilitating reliable CD.

Dual Attention-Based Global-Local Feature Extraction Network for Unsupervised Change Detection in PolSAR Images

Dual Attention-Based Global-Local Feature Extraction Network for Unsupervised Change Detection in PolSAR Images

Multi-scale Hierarchical Losses to Preserve Hidden-layer Features for Unsupervised Change Detection

Multi-scale Hierarchical Losses to Preserve Hidden-layer Features for Unsupervised Change Detection

Unsupervised flood detection on SAR time series using variational autoencoder

In this study, we propose a novel unsupervised Change Detection (CD) model to detect flood extent using Synthetic Aperture Radar (SAR) time series data. The proposed model is based on a spatiotemporal variational autoencoder, trained with reconstruction and contrastive learning techniques. The change maps are generated with a proposed novel algorithm that utilizes differences in latent feature distributions between pre-flood and post-flood data. The model is evaluated on nine different flood events by comparing the results with reference flood maps collected from the Copernicus Emergency Management Services (CEMS) and Sen1Floods11 dataset. We conducted a range of experiments and ablation studies to investigate the performance of our model. We compared the results with existing unsupervised models. The model achieved an average of 70% Intersection over Union (IoU) score which is at least 7% better than the IoU from existing unsupervised CD models. In the generalizability test, the proposed model outperformed supervised models ADS-Net (by 10% IoU) and DAUSAR (by 8% IoU), both trained on Sen1Floods11 and tested on CEMS sites. Our implementation will be available here https://github.com/RituYadav92/CLVAE-Unsupervised_Change_Detection_TimeSeriesSAR.

Unsupervised change detection in PolSAR images using siamese encoder–decoder framework based on graph-context attention network

Unsupervised change detection in PolSAR images using siamese encoder–decoder framework based on graph-context attention network