Cross-domain object detection in remote sensing suffers from substantial domain gaps arising from differences in resolution, viewing geometry, and imaging modality across sensors and platforms. Existing unsupervised domain adaptive object detection (DAOD) methods typically align source and target features using the detector’s own target-domain representations. However, the extraction of these representations is constrained by the very domain discrepancies they aim to bridge, resulting in noisy and biased features that make alignment unstable. To address this limitation, we propose the Vision-Language model Supervised Domain Adaptor (VLSDA), a domain adaptation framework supervised by a frozen vision-language model (VLM). It leverages a frozen VLM image encoder as an additional and stable semantic domain to guide domain alignment. Our VLM-supervised Prototypical Alignment (VLPA) module stabilizes category-wise alignment through a tri-domain adversarial strategy that jointly aligns source-VLM, target-VLM, and source-target distributions. Complementing this, the Global Cross-domain Contrastive Alignment (GCCA) module enhances intra-class compactness and inter-class separability via supervised contrastive learning. Without requiring any fine-tuning of the VLM, our framework directly mitigates reliance on noisy target features and improves robustness to large distribution shifts. Extensive experiments on multiple cross-domain remote sensing benchmarks demonstrate consistent improvements over state-of-the-art methods, including 68.3% mAP50on xView→DOTA and 70.5% on HRRSD → SSDD. The code is available at https://github.com/JunhongLu0704/VLSDA.

Plane and ship targets in satellite stereo images has poor stereo matching effect due to complex structure and scarcity of samples. To solve this problem, inspired by the similar representation of the same target in optical images and disparity maps, combined with prior knowledge of different components of plane and ship targets, this paper proposes a stereo matching method for plane and ship targets based on component constraint. Firstly, a generalized Hough transform segmentation method with hit ratio and uniformity ratio constraints is constructed to extract component blocks. Then, the matching cost is matched using the component block constraint. Finally, the disparity is determined by iteratively optimizing the matching cost. The experimental results show that the proposed stereo matching method NMAD is 1.4m, and RMSE is better than 5m. More advanced performance than other methods.

To address the challenge of high computational cost on transformer-based network for change detection (CD) with high-resolution remote sensing images, a neighborhood attention transformer (NAT) based on multi-scale feature fusion method (NATCD) is proposed. Initially, to effectively extract neighborhood features while reducing model complexity, a hierarchical NAT encoder is constructed for multi-scale feature extraction of bi-temporal remote sensing images. Secondly, to associate multi-scale features and alleviate the issue of poor inter-neighborhood feature correlation caused by neighborhood attention operations, a feature fusion decoder is constructed for predicting binary change maps. Experiment on LEVIR-CD and WHU-CD datasets show that NATCD achieves a better performance with a significantly computational cost than previous methods.

Considering the limitations of stereo satellite resources and imaging quality, as well as the challenges of occlusion and disparity discontinuity impacting the matching results, this paper proposes a novel generalized stereo matching method based on topological structure consistency for urban 3D reconstruction. Firstly, the topological structure within the corresponding superpixel neighborhood of the left and right views is constructed. Then, the topological structure consistency cost (TSC) is proposed to evaluate the similarity of the topological structure by combining two measurement methods of grayscale spatial distance and spatial relative relationship. Finally, iteratively optimizing the matching cost through visibility term and disparity discontinuity term to output the disparity map. The experimental results show that the proposed stereo matching method achieves normalized median absolute deviation (NMAD) better than 1.5 m and root mean square error (RMSE) better than 3.5 m, which can realize more advanced performance compared with the state-of-the-art methods.

The majority of existing object detection models struggle to fully exploit the intimate relationship between scene context and objects, and the feature fusion and proposal generation strategies tend to be relatively basic, resulting in poor model performance. To address these issues, we propose an object detection model based on adaptive multilevel fusion refinement network. Firstly, we propose an adaptive gated fusion network that dynamically assesses correlations between objects and scene information, generating a gated feature map to guide feature fusion and extract discriminative joint object-context features. Next, a proposal refinement model is proposed. By utilizing a learnable correlation-weighted coefficient, this model effectively merges low-level features with joint features, thereby mitigating spatial information deficits. We also propose an adaptive multidimensional offset strategy, which minimizes the impact of regression deviations on proposal quality by combining information offsets and spatial offsets. To optimize all subtasks, a novel multi-task loss function is proposed. Evaluated on the DOTA and HRSC2016 datasets shows that our method is superior to compared methods in object detection and harvests 77.26% and 90.68% mean average precision, respectively.

Effective feature dimension reduction (DR) from high-dimensional remote sensing images has been a significant challenge for remote sensing object recognition. Directly adopting vector-based DR method ignores remote sensing data’s inherent tensor structure information, leading to the undersample problem (USP). In addition, the existing tensor-based DR methods either require an exponential storage space increasing with the orders of the input tensor (i.e., Tucker-form methods) or are dependent on the permutation of tensor modes limiting the discriminant capability of the DR results (i.e., tensor train (TT) form methods). To conquer these problems, unlike the existing Tucker or TT form feature representation, the novel tensor ring (TR) subspace learning theory is proposed systematically and rigorously to extend the traditional vector and tensor subspace learning to the TR subspace. Then, by embedding the Fisher criterion into TR subspace, the TR discriminant analysis (TRDA) is proposed to achieve DR for remote sensing tensors with flexible tensor rank and lower storage cost. To train TRDA under different computing resources, nonrecursive and exact TRDA training methods are presented to obtain the global suboptimal and local optimal solutions, respectively. Furthermore, to adapt to the case of multisource data and unlabeled data, the multiple TRDA (MTRDA) and semi-supervised TRDA (S-TRDA) are further proposed to refine multisource features in multiple TR subspaces and absorb useful information using adaptive scatter tensor, respectively. Using optical, hyperspectral, and SAR datasets, experimental results demonstrate that the proposed TRDA can obtain better recognition accuracy and smaller storage cost than the typical vector and tensor-based DR methods.

Infrared small target detection has significant applications in airspace surveillance, warning systems and so on. Detection of infrared small targets on a single image frame is a challenging work. First of all, compared to generic object detection, infrared small targets have the characteristics of low signal-to-clutter ratio, various sizes, and shapes. In addition, false alarm sources in the complex environment also impede precise detection. In the past years, scholars have performed many studies on infrared small target detection. The majority of traditional single-frame-based methods rely on the gray-scale discontinuity between the target and the neighborhood to detect infrared small targets. However, these traditional methods rely on prior assumptions. And when the real scenarios change dramatically, it is difficult to use fixed hyperparameters to handle such variations. As we all know, the deep convolutional neural network (CNN) can automatically learn the hierarchical features of images. However, CNN-based methods could not be directly applied for infrared small targets which are even small to several pixels because the pooling layer in the networks may result in the loss of targets.

Multisource remote sensing data observed from sensors with different resolutions and physical properties will present heterogeneous tensor structures and diverse feature distributions, thus posing a significant challenge for building an effective classifier for cross-source object recognition. The representative support tensor machine classifier can inherently preserve tensor structure information of remote sensing data and obtain effective recognition ability, while it can only handle same-source and same-distributed homogeneous data and fail to deal with cross-source heterogeneous remote sensing data with complex structures and various distributions. Therefore, the domain adaptation support tensor machine (DA-STM) is proposed to learn a uniform model for cross-source object recognition. To process heterogeneous tensor from different sources, multiple factor matrices with different modes are constructed to eliminate the structural differences and reduce distribution discrepancies for multisource heterogeneous data. To excavate shared classification information across sources, the shared core tensor is established to learn the classification hyperplane jointly using multisource data, and the adaptive sample labels are then embedded into the model to recover the class information during model training. To ensure efficient training, the decomposition algorithm is developed to accelerate the solving of dual problem of DA-STM. In addition, to improve classification performance as the acquirement of sequential samples, the proposed DA-STM is further upgraded to an online version to update the classification parameters dynamically. Using multi-resolution and multi-angle optical images as well as multi-angle SAR images, experimental results demonstrate that the proposed DA-STM can obtain better recognition results than typical domain adaptation methods.

Addressing the challenge of multi-class change detection with a small sample size, a change vector analysis with adaptive weight Gaussian mixture model (CVA-AWGMM) is proposed in this article. Initially, in order to avoid errors caused by geometric distortion and imprecise alignment, we employed a neighborhood search approach when calculating the different map. Instead of direct pixel-to-pixel comparisons, we compensated corresponding pixels in the images before and after the change by finding the minimum difference within a specified pixel range. Following this, we employ the fundamental framework of CVA to extract magnitude and angle features from the change vectors, facilitating the identification of both changed and unchanged regions through threshold segmentation. Based on the above binary change detection result, a semi-supervised Gaussian mixture model is used for further change class differentiation. Recognizing the inherent challenges of effectively training classifier model with a small sample size, the clustering features in a large number of unlabeled samples and the supervised information from a few labeled samples are simultaneously utilized to co-construct the objective function of the model. Meanwhile, considering the complementary properties of magnitude and angle features, the labeled samples are used to adaptively weight the features of both to further improve the accuracy of the method. Experiments were conducted on two public data sets and one self-made data set, and the results demonstrate that the proposed CVA-AWGMM outperforms several typical methods.

Low-quality digital surface model (DSM) and blurred textures result in poor visualization of building reconstructions. In this paper, an edge-based block decomposition method for building rooftop reconstruction and side walls generation is proposed, which is applied in three-dimensional (3D) reconstruction texture mapping of buildings. The rooftop is decomposed into blocks according to the contour of the building's rooftop in the orthophoto. Each block is combined with the DSM generated by the stereo pair of satellite images to reconstruct the rooftop and interpolate to generate the side walls to obtain a regularized 3D point cloud of the building. For better texture mapping effect, single image super-resolution (SISR) method is used to enhance the texture details of remote sensing images. Experimental results show that the proposed method has better visualization effect than other 3D reconstruction methods based on satellite data.

Most of the existing saliency detection methods are affected by the complex background and weak contrast in remote sensing images (RSIs), which easily leads to confusion between salient object and background. To solve this problem, we present a general space–frequency joint saliency detection method based on spatial contrast analysis (SCA) and adaptive spectrum analysis (ASA). The proposed SCA constructs a spatial distribution function to give higher saliency to bright elements with similar appearance and compact distribution, which can avoid the interference of similar elements in the background. In addition, for the weak contrast-level salient object, an adaptive spectral energy function is proposed in ASA by combining regional dispersion and frequency spectrum characteristic, which can make full use of frequency spectrum and obtains the complete information of salient object. Finally, the SCA and ASA information are fused by a joint optimization module, which is proposed to integrate the saliency of different contrast levels objects; thus, the final saliency maps are obtained. Evaluated on the extended optical remote sensing saliency detection (EORSSD) dataset and our dataset, the proposed method is superior to other five classical methods in RSI saliency detection.

The canopy vertical structure, especially the average height, has always been regarded as an important factor in monitoring forest changes. The coherent scattering model links forest canopy information with radar observations and the random volume over ground (RVoG) model has been extensively applied to the polarimetric SAR interferometry (PolInSAR) data since it was proposed. The complex coherence of the RVoG model was originally derived in a simplified way by neglecting some factors due to the complexity of the scattering process. Thence, this letter proposed a slope three-layer scattering (STLS) model for forest parameters’ estimation in sloping mountain forest region. This model separates the vertical structure of the forest into three layers: the ground layer, the tree-trunk layer, and the canopy layer which account for the simultaneous effects of three scattering components on complex coherence. Moreover, it also corrects the distortion caused by the local terrain slope. The STLS model provides a better understanding of the microwave scattering process in the terrain slope area compared with the traditional RVoG model, S-RVoG model, and general three-layer scattering model (GTLSM) model. Finally, the STLS model has been quantitatively tested with the simulated PolInSAR data with different terrain slopes from PolSARProSim software and qualitatively tested with the spaceborne SIR-C data in Tian-Shan Mount area. The results validate the potential of the proposed STLS model in forest parameter inversion.

Multiple observation data from multisensor satellites lead to data with complex heterogeneous tensor structures, and the available object data are observed by the arbitrary combination of multisource satellites, which brings significant challenges to object recognition using typical deep learning methods, classical support vector machines (SVMs), and support tensor machines (STMs). To process multisource data represented as heterogeneous tensors effectively, the heterogeneous STM (HSTM) is proposed by integrating rank-$R$decomposition of multitensors with nu-STM to directly separate samples in heterogeneous tensor space. Furthermore, combining the shared projecting tensors used to mine related information for intercombinations and the auxiliary projecting tensors exploited to capture the individual information for intracombination, the HSTM is extended to Adaptive HSTM (AHSTM) to address the arbitrary combination of multisource data by only training once. To accelerate the training of AHSTM, the AHSTM-oriented decomposition method is proposed to use small sequential analytic optimizations instead of the original large numerical optimization. Using multiangle optical and synthetic aperture radar (SAR) satellite images, experimental results demonstrate that the proposed AHSTM obtains higher accuracy than typical SVM and STM methods for each combination of multisource data and reduces the time consumption by over 75% for the training set with different sizes compared with the typical interior-point method and the active-set method.

Considering the multiple observations from satellites with different types of sensors, the available object data are regarded as different combinations of multisource heterogeneous data. To achieve accurate object recognition, the dimension reduction (DR) technology is important to capture the low-dimensional discriminative representation containing complementary information from multisource data, while off-the-shelf DR methods can only handle input represented as vector or homogeneous tensors and fail to deal with multisource data represented as heterogeneous tensors. In addition, the existing DR methods are classifier-independent, making it difficult to ensure effective recognition under a specific classifier. To solve these problems, the DR method for Heterogeneous Tensors based on Graph-based and Classifier-oriented Embedding (HTGCE) method is proposed to learn the low-dimensional representation of multisource data using different combinations of samples. First, self-learning adjacency matrices are constructed to capture the local structure of different combinations of multisource data autonomously. Then, unlike the classifier-independent discriminant term used in existing DR methods, a classifier-oriented discriminant term is constructed to enhance the specific classifier-based recognition results. Furthermore, the reconstruction error minimization term is created to enable DR results to inherit the main information of the original data. Moreover, an adaptive weight factor is built to balance the importance of different sources for object recognition. Finally, an alternative optimization strategy is presented to solve the optimization problem of HTGCE. Using the multiresolution multiangle optical dataset and the paired optical and SAR dataset, the experimental results demonstrate that the HTGCE outperforms typical vector- and tensor-based DR methods in terms of recognition accuracy.

In the above article[1], some errors appear in equations due to incorrect fonts. First, the matrix variables should be denoted by uppercase boldface letters, e.g.,$\boldsymbol{U}_{l}^{m}$and$\boldsymbol{S}_{g}$. The vector variables should be denoted as lowercase boldface letters, e.g.,$c$and$c^{\prime}$.

Due to the capability of fast multiparametric quantitative imaging, magnetic resonance fingerprinting (MRF) is becoming a promising quantitative magnetic resonance imaging approach. However, the artifacts caused by the highly undersampled data acquisition lead to inaccurate estimation of the tissue parameter maps. Based on the assumption that the 3-D MRF data can be modeled as a piecewise smooth signal, with the discontinuities localized to the zero sets of a bandlimited function, we exploit the low-rank property of the structured Toeplitz matrix constructed from the Fourier measurements. In addition, we adopt the subspace projection scheme to improve the accuracy of parameter estimation. In order to efficiently solve the regularized problem, we propose an iterative two-stage algorithm, which alternately updates the k -space data and projects the space-time matrix into the dictionary space. Numerical experiments demonstrate that the proposed algorithm shows significant improvement in MRF time-series images reconstruction and can provide more accurate parameter maps over the state-of-the-art algorithms.

We present a deep sensor fusion method based on frustum point single shot multibox detector (PointSSD) for autonomous driving scenarios. The proposed method solves the problem of precision degradation in frustum PointNets (F-PointNet) caused by relying heavily on 2D detection and making insufficient use of RGB information. The method mainly consists of two subnetworks: pyramid segmentation network (PSNet) and PointSSD. The proposed PSNet uses a novel architecture capable of performing semantic segmentation on RGB information to generate high quality image semantic information. Using these image semantic information, point cloud semantic information is obtained through projection and is then fused with raw 3D spatial features by deep fusion. The fusion results are processed by PointSSD, which is proposed for classification and bounding box regression. Evaluated on the KITTI dataset, our method is superior to other methods in 3D classification and 3D localization. In addition, our method guarantees robustness to 2D false detections.

It is challenging to achieve high accuracy and rich details three-dimensional (3D) reconstruction of ground object using satellite remote sensing image (RSI). A 3D reconstruction method is proposed to address this problem based on multi-view 3D occupancy network. The domain adaptive multi-view image obtained by pretreatment is input into the depth residual network to extract perceptual features at first. The occupancy probability of objects at different coordinates in 3D space is obtained by combining spatial probability network with definition balance strategy. Then, the isosurface is extracted to obtain the 3D mesh of the object, and the reconstruction result is given scale by combining the attitude estimation and the original RSI parameter information. Experiments are carried out on a testing dataset composed of hundreds of object image pairs. The experimental results show that the proposed method for satellite RSI has better accuracy than the typical single-view and multi-view reconstruction methods.

To study a new framework for building detection from optical remote sensing image(RSI) that based on rectangle approximation and region growing. According to the geometric structures of buildings, a right angle detection method based on voting strategy is investigated to get building candidates. In order to accurately estimate the shape of buildings with non-convex properties, an algorithm is employed to reduce false positives by extracting outer corners of buildings from the detected right angles. Combining with similar spectral features of surfaces, a segmentation algorithm based on region growing is proposed to get the complete top surfaces of buildings. Using several RSIs of Florida and Nebraska areas in WorldView-3 image data, experimental results show that the proposed method achieves the precision ratio of 89.44% and recall ratio of 90.36%.

LIDAR point cloud and RGB images are often used for object detection in autonomous driving scenarios. This paper develops a multi-view version of Frustum PointNet (F-PointNet), to be called MVFP to reduce the rate of missed detection in F-PointNet by adding auxiliary bird's eye view (BEV) detection part. In processing MVFP, initial object detection results are obtained from F-PointNet by combining the RGB image and raw LIDAR point cloud. Simultaneously, raw LIDAR point cloud is encoded into BEV feature maps, from which 2D bounding boxes are predicted. In missed detection judgement, the intersection over union (IoU) is used as a criteria for the matching of preliminary object detection results from F-PointNet and BEV maps prediction results. 2D boxes belonging to missed-detected objects from BEV maps are projected to the pipeline of F-PointNet until all the objects in BEV maps find a matching detection result in the set of F-PointNet detection results. To evaluate the performance of MVFP, 3D object detection experiments are conducted on KITTI benchmark. The experiment results demonstrate that MVFP outperforms the original F-PointNet by 5% and 4% higher recall on the hard mode of pedestrian and cyclist.

Hierarchical detection from parking lot to vehicle based on visual saliency and angle estimation is proposed for the large-area remote sensing image (RSI). According to the brightness characteristics of parking lot, an accelerated bright-based saliency map (BBSM) is presented to locate the parking lot, which is also achieved by two-step location from rough to accurate. The suspected vehicle queues are then detected using the spectral residual saliency map and rotated to the horizontal direction by angle estimation of vehicle queue. Each vehicle in the suspected vehicle queues is identified by HOG features extraction and SVM classification. Using several RSIs at a spatial resolution of 0.5 meter with the size of 24000×24000 pixels, experimental results show that the proposed method outperforms two traditional methods and achieves the precision ratio of 87.1% and recall ratio of 90.2%.

A transfer learning method based on Weighted Histogram of Oriented Gradient (WHOG) features for ships identification is proposed, which uses labeled ships at different resolutions to identify fixed resolution ships. An improved HOG features called WHOG is presented, which have a better description of the contours on different types of ships. The training and the test samples at different resolutions obey different distributions. The JDA method considers probabilistic adaptation without performing spatial alignment. To solve this problem, Mapped Alignment-Joint Distribution Adaptation (MA-JDA) method is proposed. MA is utilized to map the source and target domain data to the same feature space, then JDA is utilized to perform probabilistic adaptation to improve transfer learning performance. Extensive experiments demonstrate that the superiority of WHOG features over traditional HOG features and the MA-JDA method is better than several state-of-the-art transfer learning methods.

In airplane target detection, there was the drawback of weak recognition ability for dark targets and high false alarm rate for detected targets. In order to address the problem, we proposed a detection method based on SAR and optical image feature fusion. It extracted texture, moment and backscattering characteristics from SAR images and combined with optical features. Moreover, the novel airplane edge templates incorporating SAR and optical images were created to acquire saliency map. During the process of detection, first, the saliency map and the One-Class-SVM (OCSVM) classifier were used to initially recognize the suspected airplane targets. Then, the combination features were adopted to further identify the misidentified airplane target. The experimental results showed that the Precision of the proposed method was 61.82% and the False Alarm Rate was 20%, which was better than the HIS-based detection method.

In cyber anomaly detection, if the detected target is significantly different from the predefined normal network data pattern, it is considered an outlier. However, the degree of deviation from the normal model is often difficult to determine, making it difficult to effectively identify attack cate gories that are similar to normal network data and have small sample sizes. To address this problem, we propose a novel anomaly detection method called a comparison network (C-Net), which has a double-branch structure for a neural network. Instead of learning the correspondence between sample values and labels by neural networks, the C-Net model fits the difference values between different classes of samples and learns the correspondence between the difference values and the labels. This approach avoids the process of determining the degree of difference and addresses the problem of low attack recognition rates for attack classes that are similar to normal network data and have small sample sizes. Our model is split into the auto-encoder network and the comparison component. The former is applied to compress the normal data and detected object to collect essential features and reconstruct the input part of the network. The comparison component then uses the reconstructed input to find the difference between the normal data and the detected object. According the degree of difference, the detected object is categorized as normal or an outlier. We performed experiments using a water storage dataset. Our model’s detection rate of the Complex Malicious Response Injection (CMRI) attack category reached 95.5%, while the cyber anomaly detection algorithms based on machine learning (OCSVM, K-means, simple-One-Class, etc.) could not detect the attack. For the KDDCUP99 data, our model achieved a 99.52% detection rate in the R2L category compared to a rate of 54.62% achieved by the cyber anomaly detection algorithms based on machine learning.

Inshore ship detection becomes challenging in high-resolution optical remote sensing image (RSI) because inshore ships are often incomplete and deformed due to the poor imaging condition and shadow of ship superstructure, and there are various interferences in harbor. A contour refinement and the improved generalized Hough transform (GHT)-based inshore ship detection scheme is proposed for RSI with complex harbor scenes. First, the suspected region of ships (SRS) is located in the entire RSI according to the line segments of ship body and docks. The contours in each SRS are then refined to repair the damaged ship head contour (SHC) using the convex set characteristics of ship head and subsequently reduce non-SHC by curvature filtering. In each refined SRS, equal frequency quantification instead of equal width quantification for R-Table construction and Gini coefficient-based decision criterion combining the number and distribution of votes are proposed to improve GHT (i.e., EG-GHT) and to extract SHCs as candidate targets. The false candidates are removed according to pixel proportion described by the structured binarization feature. Applying the border scoring strategy, the best candidates with the largest score among all the overlapped bounding boxes are selected as the final detection targets. Using the public RSIs with various cases, including turbid water, cloud occlusion, ships moored together, and ships with the different sizes, experimental results demonstrate the proposed scheme outperforms state-of-the-art contour-based methods and deep learning-based methods in terms of precision-recall rate and average precision, respectively.

It is challenging to distinguish the intensive parked vehicles in the parking lot for the high-resolution satellite images even at a spatial resolution of 0.5 meter. A vehicle segmentation and detection algorithm is proposed to address this problem based on the improved Mean Shift algorithm with the adaptive parameter adjustment. The suspected vehicle areas are first located using the regional growth with the seed point determined by the simplified Itti algorithm and non-maximal suppression. An adaptive parameter adjustment based Mean Shift method is presented to divide each suspected vehicle area into the suspected vehicle slices. To reduce the false alarms, feature vectors contained spectral and texture features are constructed and used in the classification based on support vector machine to distinguish between vehicles and non-vehicles. Experimental results on several remote sensing images at a spatial resolution of 0.5 meter present the accurate detection results for the intensive parked vehicles.

In order to detect airplane efficiently in large scene remote sensing images with high spatial resolution, a hierarchical detection framework is proposed according to the process of locating the target area in the downsampled image and then detecting the target in the original high-resolution image of that target region. First, we locate the airport in the downsampled original high-resolution image based on line features (midpoint coordinate and angle) clustering. A visual saliency model based on Itti model is then applied to detect airplanes candidates in the resulting airport region with the original high-resolution image. Finally, scale invariant feature transformation and support vector machine are used to classify the airplanes candidates. Experimental results indicate that the proposed method presents higher recall rate and detection speed. Meanwhile, the time cost of airport detection is much less than representative methods.

With the rapid development of sensor technology, we pay more attention to object detection for high-resolution SAR images. Besides, the traditional object detection methods which only use one SAR image to accomplish detection aren't enough appreciate for some cases that the background around the object is complex. In the paper, we propose an object detection method for high-resolution SAR images under the spatial constraints of optical images. It consists of three main steps: Establishment of the spatial relation between optical and SAR images, spatial constraints projection from optical images and detection under the spatial constraints in high-resolution SAR images. At the end of the paper, we take the inshore ship detection as an example to show that the proposed method can greatly improve the accuracy and efficiency of object detection against the traditional SAR object detection methods in conditions that background around the object is complex.

In the field of target detection in remote sensing images, lots of learning algorithms have been presented, among which support vector machine was widely utilized. However, this kind of vector represents only one pixel of a remote sensing image that ignores the spatial relationship of neighbors. Besides, with the increase of spatial resolution of remote sensing images, detail detection of targets become possible so that we gain more detailed information. Higher resolution leads to larger data volume, nevertheless, which makes processing efficiency decrease. In order to improve the situation, we present a Hierarchical Support Tensor Machine (H-STM) method, which deal images with feature tensors that remains much spatial structural information, and according to requirement, we detect if targets exist in low-resolution images with low-level STM first. While got desired result, we make use of a higher level to find out more detail information.

An efficient vehicle detection method is proposed for remote sensing images provided by Chinese GF-2 satellite with the 1-meter spatial resolution in this paper. Considering the characteristics of the vehicle on the road in GF-2 image, dark and light vehicles are firstly detected on the masked road using the semi-automatic dual-threshold technique. The residual disturbance (e.g. shadows, lane line and noise) is removed by the adjacent relation and double area filtering. Furthermore, to avoid the detected broken vehicle, an integrity processing is proposed to locate the detected broken vehicle and eliminate the repetition detection. Experimental results on several GF-2 images present the lower miss rate and false rate in vehicle detection.

Summary form only given: In this paper a lossless compression method for hyperspectral image is given. RKLT-based scheme was first presented by combining with 3D prediction, principal component selection, positive mapping followed by a range coder. The proposed method avoids the float number of coefficient which can make it much more easier to be processed on hardware. Numerical experiments show that the proposed method outperforms the state-of-the-art methods (i.e., LUT-NN, LAIS-LUT, and AI) by 13% in terms of compression ratio.

When dealing with the enormous remote sensing image, JPEG XR offers similar compression rate compared with JPEG2000, while consumes nearly the same resource as JPEG. Taking the limitation of power and computational resources onboard into consideration, FPGA-based hardware solution stands out with advantages of less resource requirement and higher speed. And the operation of JPEG XR is all in integer, so the image compression algorithm is feasible to implement. To realize high-throughput pipeline architecture for onboard image compression with low clock frequency, FPGA is adopted to perform the parallel architecture for JPEG XR lossy compression. The parallel architecture is summed up into five modules: pre-filter module, photo core transform (PCT) module, quantification module, predictive coding module and entropy coding module. Experimental results demonstrate the coding speed is 0.437 pixel per clock, speed of code stream is 41.7Mbyte/s at 100MHz clock frequency.

Hyperspectral images (HSI) contain hundreds of bands, which brings huge amount of data. In this paper, we propose a novel compression method for HSI with Tucker3 decomposition. The hyperspectral images are firstly decomposed into core tensor, and then the number of components is selected according to the Fast particle swarm optimization (FPSO). Compared to the traditional methods, the new method has excellent reconstruction quality and less computing time.

Spatial and spectral decor relations are necessary for hyper spectral data compression. The two dimensional wavelet transform based spatial transform and the Karhunen-Loève transform (KLT) based spectral transform have been employed successfully for hyper spectral data compression. In this paper a hyper spectral asymmetrical data compression is proposed as an improvement of the low complexity version of the Karhunen-Loève transform following the energy distribution in the wavelet transform domain. In the improved low complexity KLT, the computation processing of the covariance matrix is carried out on a spectral data which is extracted from the region of high energy distribution. The new method highlights the physical difference between the spatial and spectral characteristics of hyper spectral data. Experimental results show that the new method has improved significantly, not only the computation time but also has a good performance for the compressed data.

Most lossy compression methods for hyperspectral image (HSI) usually compress the data in this way that focus on preserving low frequency information. However, for some applications such as edge detection, the information which belongs to high frequency is more useful. Thus, a new pixel of interest (POI)-preserving-based HSI compression scheme is proposed. The concept of POI is proposed because some pixels are significant in preserving the main high frequency. Firstly, the POI extraction is performed by unmixing and the mixed pixels are viewed as POI, then the mask of the pixel of interest (MPI) is generated. Secondly, the compression scheme based on the POI preserving is conducted. The spatial and spectral redundancies are reduced, respectively, then a POI-lifting strategy is adopted for preserving the main high frequency information. Finally, bit allocation and encoding to the transformed HSI is performed by SPIHT_TCIRA algorithm, followed by the contextual adaptive arithmetic coder (CAAC). Experiments are implemented using the HSI acquired by the ROSIS Sensor. Results indicate that compared with the common compression method, the POI-preserving-based compression method can keep the key high-frequency information more effectively.

In this paper, a rapid compression technology for remote sensing image based on different ground scenes is proposed. According to the different textures and the corresponding information contained different ground types, compression parameter is set for JPEG2000, firstly. And then considering the weight of the information amount of the image, the proportion of the information amount of the image and the proportion of the weight of the image, the bit rate of the image are allocated and the compression of the image is realized. Finally, an algorithm based on the same pass truncation is proposed. Experimental results demonstrate the proposed method speeds up the compression algorithm with little loss of reconstruction quality.

Hyperspectral images (HSI) contain hundreds of bands, which brings huge amount of data. In this paper, a novel compression method based on optimal-truncation tucker decomposition for HSI is proposed. HSI tensor is firstly decomposed into complete core tensor. And then core tensor and factor matrices are truncated according to the optimal number of components of core tensor along each mode (NCCTEM), which is determined by the proposed criterion for the optimal NCCTEM and searching strategy. Experimental results show that the proposed method has the excellent reconstruction comparable to the traditional compression methods. Furthermore, it significantly reduces the compression and decompression time.

Advances in sensor and computer technology are revolutionizing the way that remote sensing data with hundreds or even thousands of channels for the same area on the surface of the earth is collected, managed and analyzed. In this paper, the classical Spectral Angle Mapper (SAM) algorithm, which is fit for parallel and distributed computing, is implemented by using Graphic Processing Units (GPU) and distributed cluster respectively to accelerate the computations. A quantitative performance comparison between Compute Unified Device Architecture (CUDA) and Matlab platform is given by analyzing result of different parallel architectures' implementation of the same SAM algorithm.

In this paper, a remote sensing building detection method is proposed, based on knowledge of environment. Digital surface model (DSM) and optical image are utilized for analyzing intricate environment of urban area. Partitioned terrain adjustment (PTA) method is proposed for alleviating terrain interference. Furthermore, vegetation interference is suppressed by optical image based reconfirming. Experiments results indicate that high accuracy detection could be obtained by our method.

Unmixing is an important technique to extract sub-pixel information contained in hyperspectral image. Many spectrum mixture models and unmixing algorithms have been proposed, but little of them consider unmixing as an inverse problem, which is usually ill-posedness, i.e. the uniqueness, existence and stability of solution may not be satisfied simultaneously. Traditional algorithms pay more attention to the former two conditions and neglect the last one. However, actual hyperspectral data is usually noise contaminated, that means the stability of unmixing algorithm is also crucial. Motivated by this, we propose a novel linear spectrum unmixing method based on regularizing operator. By modifying the original form of cost function with respect to linear mixture model, proposed unmixing algorithm reduces the condition number as well as sensitivity to noise of image. Taking semi-simulation hyperspectral image containing noise as test data, we proved thee performance on preserving unmixing effect of our method when unmixing image is noise contaminated.

A new scheme integrating segmentation into classification to analyze hyperspectral images is presented in this paper, particularly for images with a very few number of labels and largely adjacent spatial structures. Using pixel-wise semi-supervised support vector machine, the image is classified, and segmented by modified C-V level set in this method. Afterwards, classification and segmentation images are combined with neighborhood voting. Experiments are conducted on a 200-band AVIRIS image of the Northwestern Indiana's Indian Pine site. The integration of the spatial information from the level set segmentation provides classification images with more homogeneous regions and improves the classification accuracy, comparing to the general pixel-wise supervised and semi-supervised classification.

In this paper, the bandwidth of channels is finite in a practical video compression and transmission system, down-sampling techniques are usually adopted to reduce the rate of bit stream in the transmitter before video compression. Thus the reconstructed frames need to be inserted into the video sequences as the missing frames in the receiver after video decompression. Based on traditional temporal frame interpolation method using bidirectional motion vectors, an improved algorithm is proposed to classify the motional relationship to determine the weighting ratio between the adjacent frames, which is used to reconstruct the missing frame by interpolation.

In recent years, remote sensing images with high resolution are increasingly applied in change detection and disaster assessment. Compared with the traditional pixel-based methods, object-oriented image processing techniques have attracted more attention for high resolution images. In this paper, we aim to research the object-oriented change detection for urban area. A new multi-scale segmentation algorithm is proposed so as to obtain accurate image objects, and a pre-processing step is adopted to improve the computation efficiency. In order to testify the performance of proposed method, experiments are conducted on QuickBird images. The experimental results show that accurate image objects and changed area can be acquired in appropriate scales.

Hyperspectral images (HSI) regularly contain hundreds of bands, which are of different importance in the application. Most HSI compression methods usually deal with most bands in the same way, and they do not take the difference of different bands into consideration, which may cause the loss of important spectral information. In order to preserve the spectral information of interest for applications, a new band-of-interest (BOI)-preserving-based HSI compression method is proposed. The conception of BOI is proposed because some bands are significant in the specific applications, and BOI selection methods are chosen according to application requirements. BOI selection is first performed according to application measurements. Then, BOI information is fed into recursive bidirection prediction (RBP) and set partition in hierarchical trees (SPIHT) compression scheme which uses RBP for spectral decorrelation followed by SPIHT algorithm for coding the resulting decorrelated residual images. More bits are allocated to BOI to preserve BOI by two approaches, respectively. Compress BOI and non-BOI bands directly with low distortion and high distortion, respectively, and compress all bands with low distortion and perform a postcompression truncation. Experiments are implemented with different settings using AVIRIS images. Results indicate that the proposed two methods both can achieve excellent compression efficiency and reconstructed quality. In addition, they can improve the application effect in both material classification and target recognition. Compared with non-BOI compression algorithm, at the compression ratio of 80, the proposed methods improve the classification accuracy by 2% and target recognition accuracy by 9%.