2014 Vol. 39, No. 12
In this review，recent developments and future challenges in hyperspectral target detection are considered in relation to the two main approaches.The signal detection framework induced methods such as the structured backgrounds detector constrained energy minimization(CEM) and the unstructured backgrounds detector adaptive cosine/coherent estimation(ACE) are the classical methods of hyperspectral target detection，while advanced statistical pattern recognition and machine learning based approaches such as the kernel method and the sparse representation related algorithms are becoming the frontier topic in this area. The core concepts of these methods as well as their advantages and disadvantages are overviewed，and the future prospects of hyperspectral target detection are out lined.
Control point gross error detection is a critical step that guarantees the geometric correction accuracy of optical satellite images during automatic geometric correction.This paper focuses on comparison and analysis of the three classical gross error detection methods;data snooping，robust estimation(iteration method with variable weights) and random sample consensus(RANSAC).First,the steps of the three methods are described in detail.Next，gross error detection experiments using the three methods conducted with different gross error rates，i. e.10%，20%，30%and 60%，respectively are reported.These experimental results show that RANSAC is more robust and less sensitive to the gross error rate than data snooping and robust estimation and therefore the most appropriate method for gross error detection in automatic geometric correction.
Scattered textures in 3D model building have added a huge computational cost to the process of real-time rendering.This thesis proposes a texture optimization methodology for 3D building models based on super face. The approach firstly deconstructs the surface geometric characteristics of a 3D model，and then obtains a set of adjacent triangles whose normals are nearly consistent;using seeded region growing，then projects the set of triangles onto the largest projection plan and extracts the boundary to get the super faces. Finally，in order to achieve texture layout optimization and improve the utilization of texture space，it renders the scattered textures corresponding to the super faces into one texture through the optimization insert algorithm. The results show the approach can process three-dimensional building models correctly，significantly reducing the number of textures and the storage space required while maintaining the overall appearance of the scene.
In this paper，we propose a multiple feature based hierarchal component model for aero-plane detection on remote sensing image.In order to take the use of multiple features together with Part-based Model approach，we propose the new algorithm:first，in order to give a more clear description of the target，MKL learning method is used to combine the multiple feature extracted from the target through a liner combination procedure，then，because the object itself has structure features，we build a latent hierarchical structure model(LHSM)，at last，we combine the MKL and LHS together to form the algorithm proposed in the beginning. We also test the effects of the new algorithm by collecting the remote sensing images from ten international airports;the result shows that the new approach is worthwhile.
In order to evaluate the quality of the distorted image，it is necessary to calculate the similarity degree between the distorted image and the original image. By integrating gradient magnitude and gradient phase of image with structural similarity(SSIM)，this paper proposed a new image quality assessment model—gradient similarity(GSIM)，and the image quality assessment algorithm based on this model.Compared with the SSIM model and the Uradient-based model，this new model not only includes luminance，contrast and structure of image，but more important lies in that it adds gradient phase information on the new model. The result of experiments，through evaluating 982 distoned images in the LIVE database and 924 remote sensing images compression，shows that this new model is superior to traditional models of MSE，PSNR，SSIM and the Uradient-based model.This new model，contrast with traditional model of SSIM，can find better solutions to the problem of objective assessment on seriously distorted images inconsistent with the subjective perception，and also the problem of the mixing evaluation effectiveness relatively worse to multiple types distorted images.Therefore，this new model can truly reflect the quality of the visual perception of the distorted image with higher assessment reliability.
In this paper，we proposed a new segmentation method which integrates polarimetric，statistical distribution and geometric shape features of polarimetric SAR image based on the Fractal Network Evolution Algorithm(FNEA).Firstly，the similarity criterion of polarimetric features between adjacent objects was acquired based on the Pauli decomposition，while the similarity criterion of statictical feature was constructed via the coherency matrix Wishart distribution hypothesis. Secondly，a multi-feature integration strategy in objects merging was established，and polarimetric feature values were stretched in advance to make the heterogeneities of polarimetric，statistic distribution and shape features between adjacent objects to be at the similar level. Then，an integrated multi-feature segmentation flow was built according to the above processes. Lastly，this method was verified with RADA-RSAT-2 image of Altona and L band ESAR image of Oberpfaffenhofen，suggesting that it can effetetively reduce the speckle effects and obtain accurate segmentation results，especially in the homogeneous texture areas like farmland and lakes.
Chinese small satellite constellation(abbreviated HJ constellation)is special for environment and disaster monitoring and forecasting. The satellites HJ-1 A/B have the feature of wide coverage，which can be widely used in the ecological environment and disaster monitoring. The traditional positioning model can’t get ideal accuracy because of the CCD feature of wide coverage. It has significant error in CCD linear array direction. The reason is that the error factors changing with the field of view are not taken into count. Therefore，camera distortion model of CCD camera is considered and improved distortion model based on polynomial is used for image positioning in this thesis. The experimental results demonstrate the proposed algorithm improves the RMSE in CCD linear array direction is obviously reduced. The improved method has great significance to improve the positioning accuracyof HJ-1 CCD image and promote the use of domestic wide coverage satellite images.
Reconstruction method of dense point cloud using space-patch model(SPM) from sequence images is proposed. Considering gray consistency constraints and space geometric constraints of SPM，dense SPM are generated to approximating surface by selecting，expanding and filtering of seed points. This method is beyond the constraints of rigorous conditions of traditional dense matching of epipolar images.Experimental results show the validity of the algorithm and moreover. Kind of compartitive experiments are applied to analyzing that how iterations of expansion and filter influencing the quality of point cloud. The results show that massive 3D point cloud can be reconstructed by using repeatedly iterations of expansion and filter from high resolution images.
This research employ remotely sensed data and explorative data mining tools to study the characteristics of urban expansion in China. We used 27 metropolises in China as samples，and acquired the urban expansion data and socioeconomic statistics of these cities from 1990 to 2010. We selected the indices in terms of urban expansion rate，land use intensity，and landscape pattern，and then analyzed the change of these indices across cities using self-organizing feature map(SOFM) as a visual data mining tool. By examining the component planes generated by SOFM，we found that the cities expanded very rapidly in a low intensity way from 1990 to 2010. The area of urban land in 2010 became 3. 81 times of 1990 in average，but the intensity of urban population deceased annually 3%to 4%in average. These cities expanded more rapidly in a lower intensity way from 2000 to 2010 comparing with the former decade. The biggest cities expanded in most expansion rate in the first decade，while the relative small cities leading the expansion in the second decade.
Valve-closing analysis is used to obtain the optimal valve-closing scheme in maintenance of water distribution networks，which is significant for water supply security. A new pipeline analysis model，unit-valve graph，is presented based on the traditional node-link graph model，which has much less features and concise topology. As a pseudo-dual graph of the original node-link graph，the unit valve graph classifies the network regions isolated by valves as vertexes，and takes valves as edges. An efficient isolated unit constructing algorithm is designed based on the node-link graph，which helps to build the unit-valve graph. Further more，an efficient valve closing algorithm based on the unit valve graph is developed. Experimental results show that the new algorithm has a remarkable advantage in terms of efficiency compared with the traditional algorithm and get a correct valve closing scheme even in the case of multi point construction.
A simulated annealing based morphing of linear features is presented in this paper. For the two representations of the same feature in two different scales，characteristic points are extracted first from the small scale related data by using the constrained Delaunay triangulation，then the global optimal matching between characteristic points and the large scale linear feature’s vertexes is estimated by simulated annealing technique. The two linear features are divided into two groups of sub-segments by the matching results，each of the corresponding segment pair is interpolated using the linear interpolation method. Simulation example and actual data show that the Morphing method proposed in this paper，which taking into account the generalization operators of bend simplification，deletion，exaggeration and typification，can effectively keep the original linear feature’s structure characteristics and improve the accuracy of morphing transformation.
This paper proposes and implements a new method for typification of ditches with almost parallel pattern. In this method，azimuth relation graph of ditches is established first，by generating the maximum complete subgraph of the connected components in which ditches having almost parallel pattern are extracted then. Next，cognition distance between ditches is defined according to the lawsof visual cognition and then ditches are grouped based on the edge-cutting of the minimal spanning tree of ditches. Afterwards，ditches in each group are re-represented based on K-means+algorithm. Finally，typification result is evaluated through comparing the similarity between the contour of ditches and the variation of ditch density before and after typification. Experiment using data of Uuangzhoucity validates that the proposed method can implement the typification of ditches effectively.
Exiting parallel DEM preprocessing algorithms that do not consider parallel granularity.This paper presents a parallel DEM preprocessing algorithm with granularity control based on the analysis of the sequential algorithm proposed by Moran and Vezina(M&V algorithm).A Message Passing Interface(MPI) library is applied to implement the parallel algorithm. The parallel performance of the proposed algorithm is assessed by two gridded DEMs with different sizes on a multi-nodeLinux cluster. The application results show that the parallel M&V algorithm can complete the computing tasks when filling sinks and removing flat areas at any granuality，and it outputs an optimal granularity to achieve the best parallel performance for a given DEM dataset.
When using total least squares to solve the regression model parameter，both the traditional SVD and iteration methods do not consider that corrections of the same elements in different positions of the augmented matrix formed by the coefficient matrix and the observation vector are different. This paper proposes a new iteration method which can effectively solve the shortage problem in traditional methods, puts the same element in different positions of an augmented matrix into the same correction. This method is more compatible to the actual situation.Adjustment precision can also be improved. Finally, a concrete example was conducted to verify the feasibility and effectiveness of this method.
The linear semiparametric regression model L=BX+s+△ is a combine of classical linear model with nonparametric regression model. Based on the basic theory of the penalized least squares for the semiparametric mode，this paper presents a detailed discussion on the theory and method of hypothetical test for nonparameters in the linear semiparametric model and the hypothetical test statictics are derived and its distribution are proved. Finally the validity of the proposed theory and method of the hypothetical test for nonparameter is confimed by simulated experiment.
Considering the spherical visco-elastic structure，the post seismic deformation inversion mode had been constructed based on the visco-elastic earth model. We developed the inversion mode and inversion software package based on the post seismic fault dislocation model considering the crust stratified structure. Using the genetic algorithm to invert the different fault dislocation parameters due to strike slip fault，dip slip fault respectively and comparing to the result inverted from the homogenous dislocation model，the result shows that the inversion algorithm used here could invert the dislocation parameters from the large region using the genetic algorithm and the inversion result is better to use. The inversion mode proposed here could better invert the seismic source parameters to different types fault. There is also a important find to suggest us that we could not use the minimum of VT PV as the only rule to judge the inversion result whether good or not when the model mode is not obvious. We should find other ways to make a supplement judge.
In order to achieve the ambition of VOCE，a Fortran program is compiled for recovering Earth’s gravitational field model. A filter approach is designed for gravity gradients based on Fourier series.Several gravity field models are recovered by using VOCE PKI orbits and gravity gradients.The results show that the recovering ability of VOCE orbits is up to 120 degree and order. The gravity gradients are influenced by polar gaps more serious than orbits. An Earth’s gravitational field mode1 up to degree and order 200 named SWJTU2013U0 has been resolved by combining 70 days VOCE orbits and gravity gradients between 2009-11-02一2010-01-10.The precision of the model is well，but slightly less than the precision of GO_CONS_GCF_2_TIM_R3 as shown by internal and external valilotion.
Because of the Apollo lunar seismic network merely located on the near side of the Moon and the traditional methods failing in the best solutions，several lunar parameters such as load ratio f，crustal thickness bc，subsurface load depth Zb，crustal density pc，lithospheric elastic thickness Te can be solved through lunar gravity/topography admittance with PSO. Results indicate that surface and subsurface load models are rational and all the best solutions of the lunar parameters studied can be simultaneously calculated. It can provide a reference for lunar internal structure research，when such method investigated here applied to lunar geophysical parameters inversion.
A method for extracting seabed feature parameters is proposed based on the angular response curve of the multibeam backscatter strength. Based on the change rule of backscatter strength for one ping，the determining method of the number of the pings has been put forward which was inchided in the obtaining of the angular response curve of the mean backscatter strength. The relationship between angular response curve and seabed is analyaed，and using norrleast squares fitting method combine Hellequin parameters model to extract seabed feature parameters. The results show that the extracted feature parameters can suppress backscatter strength error and reserve change information of backscatter strength with incidence angle，and improve capability and reliability of multibeam seabed classification.
By decomposing the color face image into HSI space;generating a three channel face image template;and using HSI-PCNN algorithm of three channels extracts facial feature sequence，the feature sequence can be used for face recognition. An experimental comparison shows that this method can distinguish facial features well，while retaining the skin color information from the face. The HSTPCNN face recognition method，when compared to the original PCNN method performed as well as other face recognition methods. The proposed method for face detection process is simple and the identification accuracy is high. In order to achieve better outputs，after repeated experiments and parameter adjustments，this proposed method using for face recognition showed strong robustness.
A novel unsupervised color image segmentation method using graph cuts with multiple components is proposed，which can overcome the problem of the higher computational complexity caused by more labels during inferring by graph cuts. First，the quaternion cut-off window feature and CIE Lab color feature of a given image are extracted and fused based on the gradient information of the image. Then the segmentation is formulated as a labeling problem and solved by an iterative process based on graph cuts and maximum likelihood(ML) estimation. At each iteration，the connected regions in a segment are handled as sub-components of the segment instead of relabeling them with unique labels. In doing so，the number of labels does not increase，and thus the computational complexity can be reduced during inference by graph cuts. Finally，the segmentation result is obtained after removing some weak edges. Experimental results and theoretical proof demonstrate the good performance of the proposed method.