3D semantic labeling is a fundamental task in airborne laser scanning (ALS) point clouds processing. The complexity of observed scenes and the irregularity of point distributions make this task quite challenging. Existing methods rely on a large number of features for the LiDAR points and the interaction of neighboring points, but cannot exploit the potential of them. In this paper, a convolutional neural network (CNN) based method that extracts the high-level representation of features is used. A point-based feature image-generation method is proposed that transforms the 3D neighborhood features of a point into a 2D image. First, for each point in the ALS data, the local geometric features, global geometric features and full-waveform features of its neighboring points within a window are extracted and transformed into an image. Then, the feature images are treated as the input of a CNN model for a 3D semantic labeling task. Finally, to allow performance comparisons with existing approaches, we evaluate our framework on the publicly available datasets provided by the International Society for Photogrammetry and Remote Sensing Working Groups II/4 (ISPRS WG II/4) benchmark tests on 3D labeling. The experiment results achieve 82.3% overall accuracy, which is the best among all considered methods.

Scene classification is an active research area in the remote sensing (RS) domain. Some categories of RS scenes, such as medium residential and dense residential scenes, would contain the same type of geographical objects but have various spatial distributions among these objects. The adjacency and disjointness relationships among geographical objects are normally neglected by existing RS scene classification methods using convolutional neural networks (CNNs). In this study, a multi-output network (MopNet) combining a graph neural network (GNN) and a CNN is proposed for RS scene classification with a joint loss. In a candidate RS image for scene classification, superpixel regions are constructed through image segmentation and are represented as graph nodes, while graph edges between nodes are created according to the spatial adjacency among corresponding superpixel regions. A training strategy of a jointly learning CNN and GNN is adopted in the MopNet. Through the message propagation mechanism of MopNet, spatial and topological relationships imbedded in the edges of graphs are employed. The parameters of the CNN and GNN in MopNet are updated simultaneously with the guidance of a joint loss via the backpropagation mechanism. Experimental results on the OPTIMAL-31 and aerial image dataset (AID) datasets show that the proposed MopNet combining a graph convolutional network (GCN) or graph attention network (GAT) and ResNet50 achieves state-of-the-art accuracy. The overall accuracy obtained on OPTIMAL-31 is 96.06% and those on AID are 95.53% and 97.11% under training ratios of 20% and 50%, respectively. Spatial and topological relationships imbedded in RS images are helpful for improving the performance of scene classification.

With the widespread availability of satellite data, a single region can be described using multi-source and multi-temporal remote sensing data, such as high-resolution (HR) optical imagery, synthetic aperture radar (SAR) imagery, and space-borne laser altimetry data. These have become the main source of data for geopositioning. However, due to the limitation of the direct geometric accuracy of HR optical imagery and the effect of the small intersection angle of HR optical imagery in stereo pair orientation, the geometric accuracy of HR optical imagery cannot meet the requirements for geopositioning without ground control points (GCPs), especially in uninhabited areas, such as forests, plateaus, or deserts. Without satellite attitude error, SAR usually provides higher geometric accuracy than optical satellites. Space-borne laser altimetry technology can collect global laser footprints with high altitude accuracy. Therefore, this paper presents a geometric accuracy improvement method for HR optical satellite remote sensing imagery combining multi-temporal SAR Imagery and GLAS data without GCPs. Based on the imaging mechanism, the differences in the weight matrix determination of the HR optical imagery and SAR imagery were analyzed. The laser altimetry data with high altitude accuracy were selected and applied as height control point in combined geopositioning. To validate the combined geopositioning approach, GaoFen2 (GF2) optical imagery, GaoFen6 (GF6) optical imagery, GaoFen3 (GF3) SAR imagery, and the Geoscience Laser Altimeter System (GLAS) footprint were tested. The experimental results show that the proposed model can be effectively applied to combined geopositioning to improve the geometric accuracy of HR optical imagery. Moreover, we found that the distribution and weight matrix determination of SAR images and the distribution of GLAS footprints are the crucial factors influencing geometric accuracy. Combined geopositioning using multi-source remote sensing data can achieve a plane accuracy of 1.587 m and an altitude accuracy of 1.985 m, which is similar to the geometric accuracy of geopositioning of GF2 with GCPs.

Keywords: Satellite Jitter Detection, Error Compensation, Band-to-band Registration, Accuracy Evaluation, Multispectral Image Satellite jitter has become a more and more important factor which affects the quality of imagery products with development of the high resolution satellite. This paper focused on analyzing the impact on multi-spectral image caused by satellite jitter and proposed a jitter detection and compensation method to improve the band-to-band registration efficiently when jitter exists without observation by attitude sensor. Firstly, the design of multi-spectral camera and the mainstream band-to-band registration method is introduced to explain factors influencing the registration accuracy. As one of factors, satellite jitter is an unexpected satellite movement and do have impact on registration on both across and along track but easy to be ignored for the lack of high frequency and accuracy attitude data. So next the jitter detection and compensation method is proposed, in which there are six main steps to achieve the analysis of registration accuracy with and without jitter and improvement of registration accuracy after compensation when the jitter cannot be ignored. Finally, three sets of multi-spectral images of ZY-3 were used to verify the proposed method. As a result, the error caused by satellite jitter was suppressed efficiently from 0.2pixels to 0.02pixels and registration accuracy (RMSE) was improved from 0.32 pixels to 0.11 pixels by the proposed method. The results indicate that the proposed method can detect and compensate the distortion of multi-spectral image caused by satellite jitter accurately and efficiently.

In this paper, a feature line based method for building detection and reconstruction from oblique airborne imagery is presented. With the development of Multi-View Stereo technology, increasing photogrammetric softwares are provided to generate textured meshes from oblique airborne imagery. However, errors in image matching and mesh segmentation lead to the low geometrical accuracy of building models, especially at building boundaries. To simplify massive meshes and construct accurate 3D building models, we integrate multi-view images and meshes by using feature lines, in which contour lines are used for building detection and straight skeleton for building reconstruction. Firstly, through the contour clustering method, buildings can be quickly and robustly detected from meshes. Then, a feature preserving mesh segmentation method is applied to accurately extract 3D straight skeleton from meshes. Finally, straight feature lines derived from multi-view images are used to rectify inaccurate parts of 3D straight skeleton of buildings. Therefore, low quality model can be refined by the accuracy improvement of mesh feature lines and rectification with feature lines of multi-view images. The test dataset in Zürich is provided by ISPRS/EuroSDR initiative Benchmark on High Density Image Matching for DSM Computation. The experiments reveal that the proposed method can obtain convincing and high quality 3D building models from oblique airborne imagery.

The shadow-mapping and ray-tracing algorithms are the two popular approaches used in visibility handling for multi-view based texture reconstruction. Visibility testing based on the two algorithms needs a user-defined bias to reduce computation error. However, a constant bias does not work for every part of a geometry. Therefore, the accuracy of the two algorithms is limited. In this paper, we propose a high-precision graphics pipeline-based visibility classification (GPVC) method without introducing a bias. The method consists of two stages. In the first stage, a shader-based rendering is designed in the fixed graphics pipeline to generate initial visibility maps (IVMs). In the second stage, two algorithms, namely, lazy-projection coverage correction (LPCC) and hierarchical iterative vertex-edge-region sampling (HIVERS), are proposed to classify visible primitives into fully visible or partially visible primitives. The proposed method can be easily implemented in the graphics pipeline to achieve parallel acceleration. With respect to efficiency, the proposed method outperforms the bias-based methods. With respect to accuracy, the proposed method can theoretically reach a value of 100%. Compared with available libraries and software, the textured model based on our method is smoother with less distortion and dislocation.