An urban functional zone (UFZ) serves as the planning and implementation unit in urban development and management strategies. Previous works on multi-modal UFZ representation learning have integrated socio-economic attributes from points-of-interest (POIs) with visual features from remote sensing images. However, the inherent sampling bias and spatial inequality in POIs can impede the model’s discriminative capacity. To address the problems of insufficient data coverage and incomplete representation of physical-semantic sensing, we propose an interconnected, consistent and scalable framework within the physical-spatial-semantic representation space that we term TUF-Sensing. TUF-Sensing models building footprints and POIs as graph nodes, respectively, and applies a symmetrical graph convolutional architecture to capture the topology of the constructed graph, and the neighborhood influence between entities. To enhance the expressivity of nodes and stimulate neighborhood aggregation, the input features of buildings and POIs are constructed differently, using the polygonal attributes of buildings and one-hot encoding that reflects the categorical identity of POIs. The conducted experiments compared the performance of TUF-Sensing and six other methods on different scales of grids and blocks in Wuhan, China. The results demonstrate that TUF-Sensing yields significant improvements in both probability distribution- and categorical performance-based metrics, indicating its adaptability in large-scale and fine-grained UFZ recognition.

Modern deep neural networks (DNNs) use supervised learning to model the probability that a sample belongs to its label (termed positive learning), achieving human-level performance in image classification tasks. However, when encountering noisy labels, this process can lead the network to fit erroneous information, diminishing its generalization capability. Scene classification in high-resolution remote sensing images (SC-HRSI) holds considerable potential for diverse applications, also has the challenge of noisy labels when collecting large-scale datasets. Negative learning (NL), where DNNs are trained using complementary labels of samples, is noise tolerant, but often struggles to achieve adequate learning performance on practical challenging datasets. In this letter, we develop an enhanced NL (ENL) module utilizes the underlying sample distribution to refine model learning in the presence of mislabeled data, improving robustness against noise. Concurrently, we propose a selective positive learning (SPL) module, which uses dynamic class-wise confidence scores to filter likely-to-be-clean data, guiding the network toward efficient convergence. Finally, the integrated negative and positive learning (INPL) framework combines these two modules into a unified training pipeline, enhancing the model’s generalization ability and accelerating the convergence process. Experimental results on two widely used SC-HRSI datasets demonstrate the effectiveness of the proposed method compared to state-of-the-art noise robust methods.

Military conflicts have a significant impact on national security and the ecological environment. Effective detection methods for ground-moving objects in complex scenarios can be further utilized to assess damage and provide recommendations for security and environmental restoration. Current ground-moving object detection in optical remote sensing imagery struggles with balancing detection accuracy and real-time performance, hindering timely threat assessment. To address this, the study proposes ground-moving object detector (GMODet), a real-time detection method incorporating region awareness and semantic-spatial interaction to enhance the detection of partially occluded and fine-grained objects in complex environments. The framework includes three modules: the region awareness module (RAW), cross-scale context-aware feature aggregator (CCFA), and semantic-spatial progressive interaction module (SPIM), focusing on extracting discriminative features for contextual, multiscale, and semantic-spatial information. A new dataset, ground-based moving object dataset (GMOD), is constructed with four object types and high scene complexity, alongside experiments on the publicly available military vehicle remote sensing dataset (MVRSD). GMODet achieves the state-of-the-art performance, with mAP50, mAP75, and mAP scores of 65.5%, 48.5%, and 42.3% on the GMOD, outperforming the second-best results by 1.9%, 5.1%, and 1.5%, respectively. On the MVRSD, it achieves mAP50, mAP75, and mAP scores of 88.2%, 75.2%, and 61.7%, respectively. Notably, with an inference time of just 25 s on large-scale images ($9152\times 9152$pixels), GMODet showcases outstanding accuracy, speed, robustness, and generalization in ground-moving object detection.

Accurate and regularly updated global road maps have many applications in intelligent navigation and urban planning. Deep learning algorithms have recently shown promising road extraction results using various Earth observation data. However, limitations in sample annotation and domain discrepancies between different regions and sensors pose challenges to the applicability of existing road extraction methods for large-scale tasks. In this paper, we propose a novel GlobalRoadMapper scheme to extract global-scale roads at the submeter level from optical remote sensing images that only require limited labeled samples. The proposed GlobalRoadMapper is designed as a two-stage method integrating Supervised Domain Incremental and Unsupervised Domain Adaptation. The supervised domain incremental stage learns road features from the source domain. To deal with catastrophic forgetting when deep models learn new knowledge from new domains, a strategy that couples pre-domain replay and mean-teacher is proposed. The unsupervised domain adaptation stage expands knowledge learned from the source domain to unseen scenes. Multi-level domain adaptation is proposed at the image, feature, and prediction levels in the second stage to address the issue of weak cross-domain generalization. The effectiveness of GlobalRoadMapper was tested at 43 sites worldwide. Qualitative and quantitative results demonstrate that GlobalRoadMapper outperforms the existing methods for global-scale road extraction tasks. Furthermore, city-scale sub-meter road mapping was conducted in six cities from different regions worldwide. GlobalRoadMapper achieved visual results comparable to the manually annotated OpenStreetMap road networks. Overall, GlobalRoadMapper holds great potential for large-scale road extraction and can be adapted to create easily updatable road maps globally.

Urban function mapping serves as a vital role in urban management and planning tasks. To generate fine-grained recognition at spatial and semantic scale, a contrastive manner integrating comprehensive representation from multi-modal descriptions of urban functional zones (UFZs)is proposed. Abstract physical features from VHR images are obtained from the founder deep convolutional model. Spatial pattern and semantic features are extracted from geographical entities including urban buildings and POIs, respectively. The proposed model is validated in the downtown Wuhan, China, where resources and sensing data citywide are concentrated. Rich information is provided but also the challenges due to the high complexity are posed for urban functions recognition. The superior performances demonstrate that the multidimensional, especially the integration with spatial pattern of primary urban materials, enhances the exact and robust recognition on sophisticated functions of urban land.

Satellite video multi-object tracking holds significant importance in national defense security, emergency rescue, urban planning, and various other applications. This task is particularly challenging due to the complex background of remote sensing images and the dynamic nature of spatiotemporal processes. Currently, satellite video object tracking faces issues like erroneous detection and tracking, especially in cases of low contrast and missed detection and tracking for small objects. To address these challenges, this study introduces a method for multi-object tracking in satellite videos that considers both global information and local features of objects. The proposed method also incorporates weak feature enhancement techniques, specifically targeting issues such as low contrast. Extensive experimental verification on Jilin-1 satellite video data has been conducted, yielding excellent results and demonstrating the effectiveness of the proposed approach.

Regularized vector representation of artificial buildings serves as foundation of remote sensing and GIS analysis tasks. To generate exact and regular building polygons, a graph convolutional network integrating spatial morphology cognition features is proposed. Abstract image semantic features from CNN and topological spatial-cognition features from graph reasoning module are united for node features in graph construction and dynamic convolutions. The proposed model is validated on WHU building dataset and compared with Mask R-CNN and Curve-GCN. The superior performances demonstrate that the spatial cognitive information perception strategy enhances the reasonable geometric prediction on coordinates of boundary vertices.

Question answering system is an emerging information service system, which enables people to get answers easily and quickly to their questions. However, most of the existing methods do not effectively utilize semantic information and spatial properties. In this paper, a semantics-guided and spatial-aware framework for natural resources geo-analytical question answering is proposed. First, we use linguistic analysis techniques to translate natural language questions into structured texts. Then, the constructed natural resources semantic knowledge bases and query sample bases are introduced to extract the spatial and semantic information of geographic entities (i.e., precise geographic location and spatial representation). Considering most questions need to be answered based on a combination of multiple data sources, an improved user model that introduces semantic similarity is proposed to select appropriate data sources for a specific question. Finally, geo-analytical workflows are automatically generated by utilizing graph models and rule templates. Additionally, a question answering system was developed based on this framework and applied to natural resources monitoring in Hubei. The proposed framework is validated in both experiments and the case study. The results show that the proposed framework performs favorably on natural resources geo-analytical question answering tasks.

Multisource image matching is still a challenging task due to the significant nonlinear radiometric differences and scale variations. To address the problem, we present a novel phase congruency approach and hierarchical structure constraint strategy for multisource image matching. Specifically, we employ the phase congruency of image frequency domain in Gaussian scale space, and KAZE operator was used to detect feature point in the maximum moment space, which was obtained by the Fourier transform of the Log-Gabor even-symmetric filter. And then, extended phase features within the neighborhood region were generated, next feature description in the framework of polar coordinates was obtained. Finally, in the stage of image matching, we use the hierarchical structure constraint strategy for random sampling and verification. The experiments show that the proposed algorithm is superior to d2-NET, LGHD, RIFT, and other mainstream multisource image matching methods. Especially for scale change and rotation, the proposed algorithm provides a better way to describe the common features of multisource images, and results in a reliable and accurate matching for multisource image with obvious intensity and radiation differences.

Feature-based image matching is the key technology of computer vision and image understanding. Local invariant feature point matching is a classical and popular method for image matching, and the coarse error elimination method is often used to improve the accuracy of matching results. However, complex spatial transformation representation relations existed between images in the multi-object scene. The traditional coarse error elimination method is difficult to effectively remove the matching error by constructing a single transformation matrix. Thus, this paper proposes a coarse error elimination algorithm for multi-object image matching based on adaptive local space constraints. The main idea of this paper is to construct the adaptive size local space with prior knowledge of rough matching, and mine the affine constraints of the local space to remove outliers. Experimental results show the effectiveness of the proposed method, compared with six main coarse error elimination methods (AdLAM, NBCS, LLT, VFC, GLOF, RANSAC). And the proposed method can improve the matching accuracy by about 32.6% in this experiment.

Satellite videos have recently served as a new data source for a wide range of applications in traffic management and military surveillance. Due to its wider coverage, satellite videos show more advantages in large-scale monitoring than ground surveillance videos. However, pseudomotion background and low-resolution targets pose new challenges to moving vehicle detection in satellite videos, resulting in poor performance of conventional target detection methods when applied to satellite videos. To overcome this difficulty, we propose a novel moving vehicle detection approach using adaptive motion separation and difference accumulated trajectory. Specifically, a new indicator is designed to assist adaptive separation of moving targets and background, considering the scale invariance of vehicles in satellite videos. Meanwhile, we offer a vehicle discrimination algorithm based on a differential accumulated trajectory to distinguish the moving vehicles from the pseudomotion background. Experimental results on two satellite video data sets demonstrate that the proposed approach achieves better detection performance over the state-of-the-art moving vehicle detection methods.

Timely updated road networks are the basis for many real-world applications such as intelligent navigation and traffic management. Existing road updating methods based on remote sensing images learn from historical road databases to update roads. Road extraction models learned from historical images however, are not easily applied to a current image due to spectral differences; and only changed roads need updating. In this paper, an Unchanged-Guided Road Updating (UGRoadUpd) framework is proposed to improve the quality of updated road networks by limiting the road updating range and learning from historical unchanged roads. The UGRoadUpd framework identifies road changes using a novel dual-task dominant-transformer-based neural network for road change detection (DT-RoadCDNet). DT-RoadCDNet executes road segmentation and change detection simultaneously, from bi-temporal remote sensing images. The Dominant-Transformer based Global Context Modeling module in DT-RoadCDNet globally models the contextual spatial structure for improved integrity in roads and road changes. Based on the discovery of road changes, an unchanged-guided road update strategy updates the roads in changed areas by learning from the prior information provided by unchanged roads in a historical road database. Experiments on two newly annotated road change detection and update datasets confirms the effectiveness of our UGRoadUpd framework.

Abstract Acquired above variable clouds, aerial images contain the components of ground reflection and cloud effect. Due to the non‐uniformity, clouds in aerial images are even harder to remove than haze in terrestrial images. This paper proposes a divide‐and‐conquer scheme to remove the thin translucent clouds in a single RGB aerial image. Based on colour attenuation prior, we design a kind of veiling metric that indicates the local concentration of clouds effectively. By this metric, an aerial image containing thickness‐varied clouds is segmented into multiple regions. Each region is veiled by clouds of nearly‐equal concentration, and hence subject to common assumptions, such as boundary constraint on transmission. The atmospheric light in each region is estimated by the modified local colour‐line model and composed into a spatially‐varying airlight map for the entire image. Then scene transmission is estimated and further refined by a weighted ‐norm based contextual regularization. Finally, we recover ground reflection via the atmospheric scattering model. We verify our cloud removal method on a number of aerial images containing thin clouds and compare our results with classical single‐image dehazing methods and the state‐of‐the‐art learning‐based declouding method, respectively.

Vehicle re-identification supports cross-camera tracking and the location of specific vehicles in a smart city. The gallery images of vehicles are ranked based on the similarities in the appearance of objects to a vehicle query image. Previous work on vehicle re-identification has mainly focused on global or local analyses of predefined regions of vehicles to classify the vehicle images with a softmax loss function. On the one hand, separate global or predefined local regions of vehicles are often sensitive to perspective and occlusions. On the other hand, the embedding space supervised by the softmax loss function is not sufficiently compact for the object class. To solve these problems, we propose an end-to-end distance-based global and partial multi-regional deep network (DGPM) that combines multi-regional features to identify global and local differences. We exploit a three-branch architecture to learn the global and partial features from coarsely partitioned regions. A global similarity module is introduced to reduce the background information interference in the local branches. Unlike general classification, we design a distance-based classification layer that maintains consistency among criteria for similarity evaluation. Furthermore, we use spatiotemporal vehicle information to improve the vehicle re-identification results when the camera and shooting time are available. Systematic comparative evaluations performed on the large-scale VeRi and VehicleID datasets showed that our approach robustly achieved state-of-the-art performance. For instance, for the VeRi dataset, we achieve (79.39 + 2.78)% mAP and (96.19 + 2.26)% Rank-1 accuracy.

Object detection in very-high-resolution (VHR) remote sensing imagery remains a challenge. Environmental factors, such as illumination intensity and weather, reduce image quality, resulting in poor feature representation and limited detection accuracy. To enrich the feature representation and mine the underlying context information among objects, this article proposes a context-aware convolutional neural network (CA-CNN) model for object detection that includes proposal generation, context feature extraction, feature fusion, and classification. During feature extraction, we propose integrating a context-regions-of-interests (Context-RoIs) mining layer into the CNN model and extracting context features by mapping Context-RoIs mined from the foreground proposals to multilevel feature maps. Finally, the context features extracted from multilevel layers are fused into a single layer, and the proposals represented by the fused features are classified by a softmax classifier. In this article, through numerous experiments, we thoroughly explore the influence of key factors, such as Context-RoIs, different feature scales, and different spatial context window sizes. Because of the end-to-end network design approach, our proposed model simultaneously maintains high efficiency and effectiveness. We conducted all model testing on the public NWPU VHR-10 data set. The experimental results demonstrate that our proposed CA-CNN model achieves significantly improved model performance and better detection results compared with the state-of-the-art methods.

Fully convolutional network (FCN) modeling is a recently developed technique that is capable of significantly enhancing building extraction accuracy; it is an important branch of deep learning and uses advanced state-of-the-art techniques, especially with regard to building segmentation. In this paper, we present an enhanced deep convolutional encoder-decoder (DCED) network that has been customized for building extraction through the application of superpixel-based conditional random fields (SCRFs). The improved DCED network, with symmetrical dense-shortcut connection structures, is employed to establish the encoders for automatic extraction of building features. Our network's encoders and decoders are also symmetrical. To further reduce the occurrence of falsely segmented buildings, and to sharpen the buildings' boundaries, an SCRF is added to the end of the improved DCED architecture. Experimental results indicate that the proposed approach exhibits competitive quantitative and qualitative performance, effectively alleviating the salt-and-pepper phenomenon and retaining the edge structures of buildings. Compared with other state-of-the-art methods, our method demonstrably achieves the optimal final accuracies.

Synthetic aperture radar (SAR) image classification is a challenging subject due to the strong speckle noise present in SAR image processing. This paper is devoted to the unsupervised classification of single-band single-polarized synthetic aperture radar (SAR) images using multilayer level set approach. The principal concept is that we employ the gamma model to define the multilayer level set energy functional. The experiments conducted on both synthetic and real SAR images show that the proposed algorithm obtains improved experimental results in terms of both accuracy and efficiency.

To meet the demands in the rapid update of roads in basic geographic information, a novel Dominant Feature Driven Road Extraction Method (DFDREM) is proposed to extract urban roads from high spatial resolution satellite images. In the proposed DFDREM, road scenes are classified into edge-feature-dominant (EFD) roads and region-feature-dominant (RFD) roads based on directional line density at first. Then EFD roads are extracted with statistical lines structure line grouping and RFD roads are extracted with deep U-Net. Experiments on large scenes show that the proposed framework can complete the urban road network extraction task with the capability to 1) deal with interruptions caused by shadows and occlusions, 2) handle roads under construction with incomplete spectral and geometric characteristic.

Abstract This paper proposes a scale‐adaptive filtering method to improve the performance of structure‐preserving texture filtering for image smoothing. With classical texture filters, it usually is challenging to smooth texture at multiple scales while preserving salient structures in an image. We address this issue in the concept of adaptive bilateral filtering, where the scales of Gaussian range kernels are allowed to vary from pixel to pixel. Based on direction‐wise statistics, our method distinguishes texture from structure effectively, identifies appropriate scope around a pixel to be smoothed and thus infers an optimal smoothing scale for it. Filtering an image with varying‐scale kernels, the image is smoothed according to the distribution of texture adaptively. With commendable experimental results, we show that, needing less iterations, our proposed scheme boosts texture filtering performance in terms of preserving the geometric structures of multiple scales even after aggressive smoothing of the original image.

Water body extraction (WBE) has attracted considerable attention in the field of remote sensing image analysis. Herein, we present an enhanced deep convolutional encoder-decoder (DCED) network (or Deep U-Net) specifically tailored to WBE from remote sensing images by applying superpixel segmentation and conditional random fields (CRFs). First, we preclassify the entire remote sensing image into the water and nonwater areas via Deep U-Net, using the results of class membership probabilities as the unary potential in the CRF model. The pairwise potential of CRF is defined by a linear combination of Gaussian kernels, which forms a fully connected neighbor structure. Next, regional restriction is incorporated into the approach to enhance the consistency of the connected area. We use the simple linear iterative clustering algorithm to generate superpixels and correct the binary classification results by calculating their average posterior probabilities. Finally, a highly efficient approximate inference algorithm, mean-field inference, is generated for the final model. The results from the experimental application to GaoFen-2 images and WorldView-2 images demonstrate that the proposed approach exhibits competitive quantitative and qualitative performance, which effectively reduces salt-and-pepper noise and retains the edge structures of water bodies. Compared to existing state-of-the-art methods, our proposed method achieves superior final results.

Rainfall weather can always seriously deteriorate the quality of the outdoor monitoring system image. Since the decomposition based methods do not need to impose any restrictions on the types of rain, they have a wider application in removing the rain streaks. However, they still have the problems of rain residues in the low frequency component, and mis-matching the background and the rain streaks with the same gradient in the high frequency. In this condition, we propose an image rain removal algorithm based on the depth of field and sparse coding. The algorithm includes four steps: image decomposition, dictionary learning, atomic clustering based on Principal Component Analysis and Support Vector Machine, image revising based on the depth of field saliency map. Firstly, the image is decomposed by using the combination of bilateral filtering and short-time Fourier transform, so that the contour in the low-frequency part of the image can be better preserved. The depth of field saliency map of the image is utilized to eliminate the rain residues in the low frequency components, and also to solve the problem of mis-matching the background and the rain streaks with the same gradient in the high frequency components. The experimental results demonstrate that the proposed algorithm performs better both in rain removal and preserving the detailed information of the image than current methods.

Detecting geospatial targets in cluttered scenes is a profound challenge in the field of aerial and satellite image, especially some time-sensitive-targets like airplanes, ships, and cars. Most of time the problem firstly we face is how to rapidly judge whether a particular target is included in a large random remote sensing image, instead of detecting them on a given small image. In this paper, we introduced a hierarchical architecture with a coarse to fine strategy to quickly locate the targets. At the coarse layer, we used an improved saliency detection model utilizes multiple salience detection methods to quickly locate suspected regions in a large and complicated remote sensing image. Then at the fine layer with each region, without region proposal method, a single neural network predicts bounding boxes and class probabilities directly from full images in one evaluation is adopted to search small airplane objects. Unlike sliding window and region proposal-based techniques, this method is faster and more robust to target scale variation. Experimental results show the proposed method is quickly identify small targets in large-scale images.

This letter presents a novel flood detection approach using full polarimetric synthetic aperture radar (PolSAR) images based on a level set method considering prior geoinformation. The prior geoinformation includes information derived from vector data and topography data. The main approach accomplishes flood detection by the improved level set method, an active contour segmentation model, based on the classical Wishart distribution. Vector data are used to generate the zero initial level set curves. To investigate the separability between water and nonwater low-backscattering objects in PolSAR images, topography information is incorporated into the level set function as a constraint. Moreover, we introduce a piecewise statistical method to refine the result with the Kullback-Leibler divergence of circular polarization coherence. In addition, we design a new quantitative evaluation index to assess flood detection results. For validation, three real PolSAR images of flooded area are tested. The experimental results confirm the effectiveness of the proposed method.

Polarimetric features are essential to polarimetric synthetic aperture radar (PolSAR) image classification for their better physical understanding of terrain targets. The designed classifiers often achieve better performance via feature combination. However, the simply combination of polarimetric features cannot fully represent the information in PolSAR data, and the statistics of polarimetric features are not extensively studied. In this paper, we propose a joint statistical model for polarimetric features derived from the covariance matrix. The model is based on copula for multivariate distribution modeling and alpha-stable distribution for marginal probability density function estimations. We denote such model by CoAS. The proposed model has several advantages. First, the model is designed for real-valued polarimetric features, which avoids the complex matrix operations associated with the covariance and coherency matrices. Second, these features consist of amplitudes, correlation magnitudes, and phase differences between polarization channels. They efficiently encode information in PolSAR data, which lends itself to interpretability of results in the PolSAR context. Third, the CoAS model takes advantage of both copula and the alpha-stable distribution, which makes it general and flexible to construct the joint statistical model accounting for dependence between features. Finally, a supervised Markovian classification scheme based on the proposed CoAS model is presented. The classification results on several PolSAR data sets validate the efficacy of CoAS in PolSAR image modeling and classification. The proposed CoAS-based classifiers yield superior performance, especially in building areas. The overall accuracies are higher by 5%–10%, compared with other benchmark statistical model-based classification techniques.

In this paper we proposed a metric learning-based method to extract collapsed buildings from post-earthquake PolSAR imagery. In this method, eight building and orientation related features, including entropy H, the average scattering angle α, anisotropy A, the circular polarization correlation coefficient ρ and the four scattering powers of Yamaguchi 4 component decomposition with a rotation of the coherency matrix, are considered and analyzed. Then a transformation matrix is learned from collapsed and intact building samples via an improved informational-theoretic metric learning(ITML). With such a transformation matrix, the features are projected into a low-dimension space to mitigate the impact of topography and building's aspect angle. Finally a k − NN classifier is utilized to distinguish collapsed and intact buildings. The proposed method is tested on one RadarSAT-2 PolSAR image acquired after 2010 Yushu Earthquake in the Qinghai Province of China. Results are validated by the manually interpretation map of a very high resolution (VHR) optical image. It shows that, the method is efficient to extract collapsed building areas using limited samples and only one post-earthquake PolSAR image.

The classification of damaged building types has received increasing attention in recent years. The detection of damaged rooftop areas is crucial to improve the accuracy of classification of building damaged types. In this letter, an approach for the automatic detection of damaged rooftops areas based on the visual bag-of-words (BoWs) model is presented. First, the building rooftop is segmented into different superpixel areas. Then, the visual BoWs model is employed to build semantic feature vectors for damaged or nondamaged parts of each superpixel area. Finally, damaged and nondamaged parts of rooftop superpixel areas are discriminated using support vector machine. An evaluation of experimental results, for a selected study site of the Beichuan earthquake ruins, Sichuan, China, shows that this method is feasible and effective for the detection of damaged rooftop areas.

Automatic optical-to-SAR image registration is considered as a challenging problem because of the inconsistency of radiometric and geometric properties. Feature-based methods have proven to be effective; however, common features are difficult to extract and match, and the robustness of those methods strongly depends on feature extraction results. In this paper, a new method based on iterative line extraction and Voronoi integrated spectral point matching is developed. The core idea consists of three aspects: 1) An iterative procedure that combines line segment extraction and line intersections matching is proposed to avoid registration failure caused by poor feature extraction. 2) A multilevel strategy of coarse-to-fine registration is presented. The coarse registration aims to preserve main linear structures while reducing data redundancy, thus providing robust feature matching results for fine registration. 3) Voronoi diagram is introduced into spectral point matching to further enhance the matching accuracy between two sets of line intersection. Experimental results show that the proposed method improves the matching performance. Compared with previous methods, the proposed algorithm can effectively and robustly generate sufficient reliable point pairs and provide accurate registration.

This paper proposes a novel kernel change detection algorithm (KCD). The input vectors from two images of different times are mapped into a potential much higher dimensional feature space via a nonlinear mapping, which will usually increase the linear margin of change and no-change regions. Then a simple linear distance measure between two high dimensional feature vectors is defined in features space, which corresponds to the complicated nonlinear distance measure in input space. Furthermore the distance measure's dot product is expressed in the combination of kernel functions and large numbers of dot product processed in input space by combined kernel tactic, which avoids the computational load. Finally this paper takes the soft margin single-class support vector machine (SVM) to select the optimal hyper-plane with maximum margin. Preliminary results show the kernel change detection algorithm (KCD) has excellent performance in accuracy.