Classifying remote sensing images is vital for interpreting image content. Presently, scene classification methods using convolutional neural networks have drawbacks, including excessive parameters and heavy calculation costs. More efficient lightweight CNNs fewer calculations, but their performance generally weaker. We propose a more network method to improve accuracy with small training dataset. Inspired by fine-grained visual recognition, this study introduces bilinear model...

Remote sensing image scene classification is an important means for the understanding of remote images. Convolutional neural networks (CNNs) have been successfully applied to and demonstrated remarkable performance. However, with improvements in resolution, categories are becoming increasingly diverse, problems such as high intraclass diversity interclass similarity arisen. The performance ordinary CNNs at distinguishing complex images still limited. Therefore, we propose a feature fusion...

The detection and recognition of oriented objects in remote sensing images is a challenging task due to their complex backgrounds, various sizes, diverse aspect ratios, especially arbitrary orientations. Many object algorithms need obtain accurate angles or adopt anchors predict the bounding boxes. When directly predicting objects' boxes, loss angle discontinuous during training, which makes it difficult boundary objects. And also aggravate problems class imbalance computational cost. To...

Point cloud classification of airborne light detection and ranging (LiDAR) data is essential to extract geoinformation. Although deep learning provides a new approach for classification, the time-consuming training process dependence prevent its widespread application point clouds. To solve these problems leverage potential high-performing neural networks, we propose an LiDAR method based on transfer learning. A strategy generate feature images considering spatial distribution first...

A Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM)-aided geopositioning method is proposed to solve the problem of without ground control points for Mapping Satellite-1 imagery. The comprises coarse and accurate correction stages, it compensates errors gradually. DEM extraction matching are important steps in both objectives which compensate relative absolute an image, respectively. SRTM integrated into all processes take full advantage its consistent high accuracy....

In this paper, we present a new approach for tracking targets with their size and shape time-varying, based on combination of mean-shift affine structure. Although the well-known colour-based algorithm is an effective tool, difficulties arise when it applied to track size-changing visual target due fixed kernel-bandwidth. To improve this, study employs corner detector object candidate from reconstructs position relative scale between frames using structure available two or three views....

Photogrammetry based on high-resolution satellite image can acquire geospatial information within a large area rapidly and timely, but its geopositioning accuracy is highly dependent ground control points. Under the background of global mapping, public digital elevation model (DEM) assisted Chinese scheme was proposed to realize photogrammetry without To make full use DEM advantages consistent high accuracy, regarded as reference data matched with extracted from image, then determined...

With the development of remote sensing technology, source data is getting more abundant and resolution becoming higher. Consequently, conventional change detection method can't meet application requirements any more. In this paper, an object-oriented for multisource images using multi-feature fusion was proposed to solve problem. On basis objects acquisition multiple features extraction, SVM adopted its outstanding character in high dimensional classification. Through efficient combination...

Hybrid change detection (HCD) for high-resolution imagery usually adopt decision-level method and rely on artificial design. To address this issue, we propose a novel feature-level fusion strategy HCD based iterative slow feature analysis (ISFA). First, objects are obtained by multiresolution segmentation of bi-temporal images respectively, corresponding sets constructed through stacking pixel- object-level spectral features. Then, (SFA) is used transforming the into new space at first time....

Accurate roof segmentation is one of the key steps for automatically constructing three-dimensional (3-D) building models. Building roofs can differ significantly in terms their size, shape complexity, and number, rendering many existing airborne Light Detection And Ranging (LiDAR) methods ineffective. Thus, applicability precision these need to be improved. For this purpose, paper proposes a new method LiDAR point clouds. The proposed integrates novel region growing strategy RANdom SAmple...

The classification of airborne LiDAR point cloud is one the key procedure for its further processing and application. Aiming at difficulty obtaining high accuracy reducing time simultaneously, a transfer learning-based method classifying proposed. Firstly, three types low-level features, i.e. normalized height, intensity normal vector are calculated each point, by setting different size neighborhood, multi-scale feature images generated utilizing proposed image generation method. Then,...

In this paper, according to the building characteristics in high-resolution remote sensing imagery, a multi-feature and multi-scale method is proposed. First, based on imagery processing, index constructed by multi-direction gradient operators, some rectangle buildings are extracted index, morphology open operation shape features. Subsequently, voting matrix calculated number of pixels included intersection expansion results shadow determine light direction. The initial extraction completed...

The mean-shift algorithm is a robust and easy method of finding local extrema in the density distribution data set. It has been used successfully for visual tracking which target modelled using colour histogram, image window with best matching histogram sought. estimation these distributions essential since goodness convergence largely depends on them, so if accuracy can be improved, performance will also potentially improve, we would reduce risk losing from accumulation bias. However,...

2D human pose extraction has been completed by multiple frameworks, such as [1], [2], [3], [5], [6], [7], [8], [9], [1]0,[1][1] and other detectors. In this paper, after comparing various detectors, 1.7 version of openpose, the most cutting-edge CMU, is adopted detector paper for accuracy. As transformation from joint to 3D pose, Pavllo et al. [4] was referred to, while some modifications were made errors still existing in actual output results. We use classic GAN[23] model, network adopts...

The extraction of multi-view matching points is one the key elements in 3D reconstruction image scene, because results will directly affect accuracy reconstruction. With conversion from to dynamic connectivity, a solution based on Union Find algorithm was designed. efficient tree structure with parent-link used organize nodes sets, so that it only needed modify addressing parameter single node each process adding pair, which avoided computational traversal array compare and improved...