Yehua Sheng

Adaptive affinity propagation with spectral angle mapper for semi-supervised hyperspectral band selection

Band selection is a commonly used approach for dimensionality reduction in hyperspectral imagery. Affinity propagation (AP), a new clustering algorithm, is addressed in many fields, and it can be used for hyperspectral band selection. However, this algorithm cannot get a fixed number of exemplars during the message-passing procedure, which limits its uses to a great extent. This paper proposes an adaptive AP (AAP) algorithm for semi-supervised hyperspectral band selection and investigates the effectiveness of distance metrics for improving band selection. Specifically, the exemplar number determination algorithm and bisection method are addressed to improve AP procedure, and the relations between selected exemplar numbers and preferences are established. Experiments are conducted to evaluate the proposed AAP-based band selection algorithm, and the results demonstrate that the proposed method outperforms other popular methods, with lower computational cost and robust results.

Data environment construction for virtual geographic environment

Virtual geographic environment (VGE) aims to express the real-world naturally, and support the complex geographic analysis. The data environment, fundamental of VGE, is expected to support the data management, analysis, sharing and application requirements of the massive complex geographic spatio-temporal data. In this paper, we summarized the key problems in the construction of the data environment of VGE. The unified spatio-temporal data model and a new data structure were developed according to the geographic rules. The organization and compress storage mechanism of massive spatio-temporal data were also developed. With these foundations, case studies, which integrate the global, regional and city scale data to operate complex data modeling and analysis, are performed. The results showed that the construction of the integrated data environment of VGE can largely improve the efficiency of GIS analysis, which also provides a potential new tool to support the complex geographic analysis.

Automatic Detection and Recognition of Traffic Signs in Stereo Images Based on Features and Probabilistic Neural Networks

Considering the problem of automatic traffic sign detection and recognition in stereo images captured under motion conditions, a new algorithm for traffic sign detection and recognition based on features and probabilistic neural networks (PNN) is proposed in this paper. Firstly, global statistical color features of left image are computed based on statistics theory. Then for red, yellow and blue traffic signs, left image is segmented to three binary images by self-adaptive color segmentation method. Secondly, gray-value projection and shape analysis are used to confirm traffic sign regions in left image. Then stereo image matching is used to locate the homonymy traffic signs in right image. Thirdly, self-adaptive image segmentation is used to extract binary inner core shapes of detected traffic signs. One-dimensional feature vectors of inner core shapes are computed by central projection transformation. Fourthly, these vectors are input to the trained probabilistic neural networks for traffic sign recognition. Lastly, recognition results in left image are compared with recognition results in right image. If results in stereo images are identical, these results are confirmed as final recognition results. The new algorithm is applied to 220 real images of natural scenes taken by the vehicle-borne mobile photogrammetry system in Nanjing at different time. Experimental results show a detection and recognition rate of over 92%. So the algorithm is not only simple, but also reliable and high-speed on real traffic sign detection and recognition. Furthermore, it can obtain geometrical information of traffic signs at the same time of recognizing their types.

Geographic scenario: a possible foundation for further development of virtual geographic environments

ABSTRACT It has been two decades since virtual geographic environments (VGEs) were initially proposed. While relevant theories and technologies are evolving, data organization models have always been the foundation of VGE development, and they require further exploration. Based on the comprehensive consideration of the characteristics of VGEs, geographic scene is proposed to organize geographic information and data. We empirically find that geographic scene provides a suitable organization schema to support geo-visualization, geo-simulation, and geo-collaboration. To systematically investigate the concept and method of geographic scene, Geographic Scenario is proposed as the theory on developing geographic scene, and corresponding key issues of the Geographic Scenario are illustrated in this article. Prospects of the proposed method are discussed with the hope of informing future studies of VGEs.

Optimized extreme learning machine for urban land cover classification using hyperspectral imagery

This work presents a new urban land cover classification framework using the firefly algorithm (FA) optimized extreme learning machine (ELM). FA is adopted to optimize the regularization coefficient C and Gaussian kernel σ for kernel ELM. Additionally, effectiveness of spectral features derived from an FA-based band selection algorithm is studied for the proposed classification task. Three sets of hyperspectral databases were recorded using different sensors, namely HYDICE, HyMap, and AVIRIS. Our study shows that the proposed method outperforms traditional classification algorithms such as SVM and reduces computational cost significantly.

Hyperspectral image classification based on volumetric texture and dimensionality reduction

A novel approach using volumetric texture and reduced-spectral features is presented for hyperspectral image classification. Using this approach, the volumetric textural features were extracted by volumetric gray-level co-occurrence matrices (VGLCM). The spectral features were extracted by minimum estimated abundance covariance (MEAC) and linear prediction (LP)-based band selection, and a semi-supervised k-means (SKM) clustering method with deleting the worst cluster (SKMd) bandclustering algorithms. Moreover, four feature combination schemes were designed for hyperspectral image classification by using spectral and textural features. It has been proven that the proposed method using VGLCM outperforms the gray-level co-occurrence matrices (GLCM) method, and the experimental results indicate that the combination of spectral information with volumetric textural features leads to an improved classification performance in hyperspectral imagery.

Hyperspectral feature recognition based on kernel PCA and relational perspective map

A novel joint kernel principal component analysis (PCA) and relational perspective map (RPM) method called KPmapper is proposed for hyperspectral dimensionality reduction and spectral feature recognition. Kernel PCA is used to analyze hyperspectral data so that the major information corresponding to features can be better extracted. RPM is used to visualize hyperspectral data through two-dimensional (2D) maps, and it is an efficient approach to discover regularities and extract information by partitioning the data into pieces and mapping them onto a 2D space. The experimental results prove that the KPmapper algorithm can effectively obtain the intrinsic features in nonlinear high dimensional data. It is useful and impressing for dimensionality reduction and spectral feature recognition.

Feature-constrained surface reconstruction approach for point cloud data acquired with 3D laser scanner

Surface reconstruction is an important task in the field of 3d-GIS, computer aided design and computer graphics (CAD & CG), virtual simulation and so on. Based on available incremental surface reconstruction methods, a feature-constrained surface reconstruction approach for point cloud is presented. Firstly features are extracted from point cloud under the rules of curvature extremes and minimum spanning tree. By projecting local sample points to the fitted tangent planes and using extracted features to guide and constrain the process of local triangulation and surface propagation, topological relationship among sample points can be achieved. For the constructed models, a process named consistent normal adjustment and regularization is adopted to adjust normal of each face so that the correct surface model is achieved. Experiments show that the presented approach inherits the convenient implementation and high efficiency of traditional incremental surface reconstruction method, meanwhile, it avoids improper propagation of normal across sharp edges, which means the applicability of incremental surface reconstruction is greatly improved. Above all, appropriate k-neighborhood can help to recognize un-sufficient sampled areas and boundary parts, the presented approach can be used to reconstruct both open and close surfaces without additional interference.

Geometric quality assessment of trajectory-generated VGI road networks based on the symmetric arc similarity

As large amounts of trajectories from a wide variety of Volunteered Geographic Information (referred to as VGI) contributors pour into the spatial database, the geometric qualities of the VGI road networks generated from these trajectories are different from the ground truth road dataset and so need to be differently assessed. To address this issue, an assessment approach based on symmetric arc similarity is proposed, and the geometric quality of a VGI road network is assessed by its conformity with the corresponding ground truth road network, the results being visualized as hierarchical thematic maps. To compute the conformity, the geometric similarity between the VGI road arc and the corresponding ground truth road arc, which is selected by the adaptive searching distance, is measured based on the symmetric arc similarity method; the geometric quality is assessed based on an assessment matrix. Also, the symmetric arc similarity method is independent of directions and with a feature of shift-independence, which is applicable to assess the geometric qualities of different VGI road networks and makes the assessment result consistent with the actual situation of the real world. The robustness and scalability of the approach are examined using VGI road networks from different sources.

Stereo Image Matching for Vehicle-Borne Mobile Mapping System Based on Digital Parallax Model

Considering automatic and effective stereo image matching for vehicle-borne mobile mapping system (VMMS), a new stereo image matching algorithm based on digital parallax model (DPM) is proposed in this paper. The new matching propagation strategy is designed in this algorithm, which includes two processes as DPM construction and parallax prediction. With some known matched points, the DPM of stereo image pairs is firstly constructed, and parameters for confirming conjugate epipolar line is also worked out. Then searching range during dense matching can be confirmed under constraints of DPM and epipolar line, which can improve matching speed and accuracy. Furthermore, to improve matching robustness, the computation model of similarity measurement combined with local structure feature and global color feature is designed. The new algorithm is applied to actual stereo images taken by VMMS to verify its validity. Experimental results show that the proposed approach has higher reliability and accuracy.