This PDF file contains the front matter associated with SPIE Proceedings Volume 7147, including the Title Page, Copyright information, Table of Contents, Introduction (if any), and the Conference Committee listing.

Given the shortage of classified methods for remote sensing informations at present, the Self-organizing Artificial Neural Network is applied to classifying for TM image in order to improve classification accuracy in this paper. At the same time, as for the effecting factors of classification remote sensing image, Surface structure is considered as important parameter, which is different from other classified methods only considering spectral characters(including ENVI, Tasseled Cap, principle components, TM seven bands and etcs). Taking example for the research area of Guangzhou city, comparing with the traditional maximum likelihood classification, the result shows that the Self-organizing Artificial Neural Network is better than the supervised Maximum likelihood classification and the new method is more efficient. It is very important to provide one new mean for the classification of surface object characters in remote sensing image.

This paper mainly explains two methods about how to improve the accuracy of remote sensing classification. The first method is composite hierarchical classification with multiple information sources, assisted by GIS and based on statistical interpretation. The other one is analysis on background parameters. The experiment shows that they not only improve the methods of the conventional remote sensing classification, but also raise its accuracy.

The disparities of features that represent the same real world entities from disparate sources usually occur, thus the identification or matching of features is crutial to the map conflation. Motivated by the idea of identifying the same entities through integrating known information by eyes, the feature matching algorithm based on spatial similarity is proposed in this paper. Total similarity is obtained by integrating positional similarity, shape similarity and size similarity with a weighted average algorithm, then the matching entities is achieved according to the maximum total similarity. The matching of areal features is analyzed in detail. Regarding the areal feature as a whole, the proposed algorithm identifies the same areal features by their shape-center points in order to calculate their positional similarity, and shape similarity is given by the function of describing the shape, which ensures its precision not be affected by interferes and avoids the loss of shape information, furthermore the size of areal features is measured by their covered areas. Test results show the stability and reliability of the proposed algorithm, and its precision and recall are higher than other matching algorithm.

MODIS data has a high temporal and spectral resolution, and it can provide vegetation indices of high quality. By using MODIS NDVI time series with 250 m spatial resolution which were composite of 16 days in 2005, this work chose annual modulus of vector, maximum and minimum NDVI three indices to do classification. Training and validation samples were selected based on TM images and the 1:1,000,000 vegetation atlas of China. Then the land coverage map was generated using maximum likelihood classification (MLC) method. After post-classification process of the original classification result, the final land classification map of Keerqin sandy land was got in the end. The classification accuracy was assessed using validation samples and the result indicates that 250 m MODIS NDVI time series has advantage and potential in regional land coverage mapping. Also the classification method used in the paper could not only reduce the data amount and quicken the speed of classification, but also could reduce the disturbance of other invalidation information to classification and get better classification accuracy.

Through systematically analysises of existing multi-class SVMs (M-SVMs) methods, it is shown that hierarchy multi-class SVMs (H-SVMs) can be relatively effective. Further analysis shown that existing methods that measure separability between different classes are not suitable for kernel feature space. A new method is presented for separability measure in feature space based on the characters of RBF kernel function and SVMs. Based on the new separability measure, two kinds of H-SVMs, Binary Tree SVMs (BT SVMs) and Single Layer Clustering SVMs (SLC SVMs) are presented. They are both implements of following ideal: the higher a pair of two sub-classes is in the hierarchy, the easier to separate them. In this way, we can not only achieve classification accuracy by alleviate error accumulation from top to bottom, but also rise classification speed by reduce support vectors in classifier. Experimental results justify the rationality of the new separability measure and effectiveness of BT SVMs and SLC SVMs.

Uranium deposit has very important status in our national defiance and economy qua a kind of strategic mineral resources. It is important to point out which information should be extracted as ore-searching ones of uranium deposit. In this study, a kind of high spatial resolution satellite data-QuickBird satellite data, with 0.6m class panchromatic (Pan) and 2.4m multi-spectral stereoscopic data, was used to extract the ore-searching information of uranium deposit in Bashibulake area at the north of Tarim basin. By using effective methods of image processing, the information of ore-bearing bed, ore-control structure and mineralized alteration have been extracted successfully. The results show a high consistency with the field survey. The aim of this study is to explore practicability of high spatial resolution satellite data for prospecting minerals, and to broaden the thinking of ore-searching at similar areas.

In this paper, two methods for calculating urban compactness are introduced: one of them is a pixel based method designed by the author before and the other one is the improving upon the compact ratio. These two methods were successfully programmed using C++ language and developed into a useful software tool which could calculate the urban compactness directly from remote sensing classification image, showing great convenience for the researchers of studying urban compactness. At last, the software tool was used for the change analysis of urban compactness in the cities of Dongguan and Foshan. The results showed that the increase of urban compactness in Dongguan city mainly occurred in the outside part. For the city of Foshan, compactness changes mainly occurred in the first to the third cirque area.

This paper presents two new methods for segmentation of LIDAR points data by height histogram analysis. In this researcch, height information slicing is first conducted by height interval or number of dispersion, and histogram is performed by statistics of sliced height information. Then height histogram is analyzed to determine the segmentation mode. According to the different segmentation mode, certain optimal threshold algorithms are chosen and applied to histogram, and optimal threshold or series of optimal thresholds are obtained to finish the division process. The two methods, Multi-thresholds By Total Histogram and Multi-thresholds By Subarea Histogram, are discussed and implemented to a certain data set. At the end of the paper, some case studies are given to achieve the segmentation results and effects of slice number and the chosen algorithms are compared.

In this paper, both the advantage and disadvantage of traditional Genetic Algorithm and Back Propagation Algorithm are analyzed. A mixed algorithm on combination of genetic algorithm and back propagation algorithm is adopted and it's network is trained by a sample of the spectrum, texture, position and geometry features of hyperspectral remote sensing images. The numerical experiment is designed to compare GA-BP algorithm with an evolutionary BP algorithm, which is trained by Levenberg-Marquart algorithm. The results proved that GA-BP algorithm is more efficient, robust and practical.

Remotely sensed images are used for many purposes in today's world. In this paper, we explore the potential application of high resolution satellite images in extracting features and classifying urban settlements. The test area is Soweto, an urban area in the Greater Johannesburg Metropolitan area, in Gauteng, South Africa. We propose a new settlement typology for efficient classification of formal and informal settlements via QuickBird satellite images. Following on, an automated classification procedure based on the local binary pattern texture features is introduced. Using a convenience sample of 25 images, we show the feasibility of the new typology by applying it to both a manual classification procedure and an automated one. The manual classification procedure was conducted by a group of five experts who interpreted the images and classified them according to formal and informal settlements. Analysis of the results revealed an overall mean classification accuracy of 99.2% with a standard deviation of 1.79%. The automated method involved extracting tiles at random positions within the 25-sample dataset. The features extracted from these tiles were classified using a support vector machine. Classification accuracy on new samples was 56.27%, but cross-validation on the training data reached classification accuracies of 98%.

An integrated approach of ISODATA and SVR is presented to extract the objective information, e.g. wheat, which can adequately combine the advantages of both hard and soft classification. It exploits the classification method of ISODATA for the typical objective feature and SVR mixed spectral unmixing for the mixed objective feature. The accuracy assessment shows that this method, which can obtain a higher accuracy than that of either linear spectral unmixing or ISODATA method, is practical.

The purpose of this research was to evaluate six classifiers applied to Landsat-7 data for accuracy of Level II land-cover categories in Ohio. These methods consist of (1) USGS National Land Cover Data; (2) the spectral angle mapper; (3) the maximum likelihood classifier; (4) the maximum likelihood classifier with texture analysis; (5) a recently introduced hybrid artificial neural network; (6) and a recently introduced modified image segmentation and object-oriented processing classifier. The segmentation object-oriented processing (SOOP) classifier outperformed all others with an overall accuracy of 93.8% and Kappa Coefficient of 0.93. SOOP was the only classifier to have by-class producer and user accuracies of 90% or higher for all land-cover categories. A modified artificial neural network (ANN) classifier had the second highest overall accuracy of 87.6% and Kappa of 0.85. The four remaining classifiers had overall accuracies less than 85%. The SOOP classifier was applied to Landsat-7 data to perform a level II land-cover classification for the state of Ohio.

Remote sensing image classification is an important means for quantified remote sensing image analysis, and remote sensing image fusion can effectively improve the accuracy of image classification. This paper proposes a classification algorithm of remote sensing fused images based on independent component analysis (ICA), topographic independent component analysis (TICA), support vector machines (SVMs) and D-S evidence theory. Firstly a novel method of fusing panchromatic and multi-spectral remote sensing images is developed by contourlet transform which can offer a much richer set of directions and shapes than wavelet. As independent component analysis not only can effectively remove the correlation of multi-spectral images, but also can realize sparse coding of images and capture the essential edge structures and textures of images, then using features extracted from the ICA and TICA domain coefficients of the fused images, the SVMs are trained to classify the whole fused images. Finally apply the proposed novel D-S evidence combination scheme to make decision fusion for different classification results with different features obtained by SVMs. Experimental results show that the proposed algorithm can effectively improve the accuracy of image classification.

This paper present a detail processing procedure about SPOT5 image applied for vegetation type recognition, and determines the capacity of high spatial resolution satellite image data to discriminate vegetation type in a complex ecosystem. A high spatial resolution SPOT5 image, captured in April 2005, and coincident field data covering the Dagou valley, was used in this analysis. Image geometric rectification and image fusion are then introduced to take prepare for classification. Subsequently, a maximum likelihood classification algorithm was applied to the SPOT5 image data to map the vegetation classes. Field validation and accuracy assessment are crucial to ensure the reliability of classification results. The strategy of field work and the resulting accuracy evaluations were presented, and yielded the high classification accuracy (overall accuracy=83.86%, Kappa=80.23%). The result showed that the information on vegetation types can be mapped effectively from high spatial resolution satellite image data.

Knowledge of the area and distribution of cropland is important for land management and land security. Low spatial resolution imagery is one of the important remote sensing data source in the study of the large extent cropland. There exist many mixed pixels and effective method that should be improved to deal with them. In this paper, linear mixing model was used to unmix the time series of MODIS-NDVI data. The emphasis was the identification and extraction of endmembers, which represent the spectral characteristics of the single pure land cover types. A new endmembers extraction algorithm based on the temporal series of MODIS-NDVI and TM sample data was presented in this paper. We used the effective endmembers to linear spectral mixture model to achieve the wheat area in the study area. Regarding the classification of TM as the reference data, we evaluated the classification results and found wheat distribution's region accuracy and pixel accuracy reach to 92.9% and 0.837 respectively, which were higher than the clarification result based on the endmembers from MODIS-NDVI pixel purity index analysis or from classifications of TM data. This shew that our endmembers extraction algorithmwas available and effective, which helped to improve monitoring accuracy of large scope and distribution of vegetation.

Based on single-temporal MODIS data of Gansu province, mainly using its spectra information, three classifiers - the Maximum likelihood, BP neural network and decision tree based on data mining software See 5.0 are applied in the Land cover classification research. The validated results show that decision tree algorithm has the best performance of extraction with an overall accuracy of 82.13 percent, followed by the BP network algorithm, and that of the maximum likelihood classifier is worst; the accuracy of low vegetation area is improved with the indexes of TVA and TVD; Data mining software of See 5.0 with boosting technique can build decision tree quickly and improve the precision of miscible classes.

In order to overcome the deficiencies of traditional uncertainty assessment methods of remote sensing images classification by error-matrix and kappa coefficient, classification uncertainties at pixel scale of Beijing-1 small satellite multi-spectrum remote sensing images were measured and represented. Firstly, an unsupervised classification algorithm-neighborhood EM considering spatial autocorrelation and classification fuzziness-was introduced. Then, four uncertainty assessment indexes of neighborhood EM classification-fuzzy membership residual, relative maximum fuzzy membership deviation, fuzzy membership entropy and relative fuzzy membership entropy - were constructed. Finally, the experiments concerned were performed using Beijing-1 small satellite multi-spectrum remote sensing image data in Dongkunlun, Qinghai province, China.

In this paper, we present a novel approach for semi-automatic extraction of ribbon road axes from high resolution remotely sensed imagery. The core of our system is a road tracker based on T-shaped template matching. T-shaped template is composed of a profile perpendicular to the road axis and a rectangle parallel to and as wide as the road marks or strips of vegetation. Actually, the T-shaped template matching is an integration and improvement of typical profile matching and rectangular template matching. At the same time, parabola is deployed to model the road trajectory to predict the position of subsequent road points and to guide the tracking go through bad road conditions. Simultaneously, the least square matching is employed to search the precise road centerline point. Extensive experiments demonstrate that our proposed algorithm can fast and reliably trace roads with road marks or strip of vegetation.

Construction of network clusters and identifying hub nodes from networks has attracted more and more attentions in spatial network analysis. In this paper, we proposed clustering algorithm and outlying node detection algorithm for spatial road network analysis. Network clustering algorithm consists of constructing clusters and creating a simplified structure of the network. When performing clustering on the network, we introduced the definitions of strong cluster and weak cluster, where each node has more connections within the cluster than with the rest of the graph, for achieving reliable and reasonable clusters. For users' understanding the structure of the network, we constructed a simplified graph approximation of the network, whose nodes were representative nodes in clusters of the network, and edges were the connections between those representative nodes. In outlying node detection algorithm, a node is identified as an outlier, not because of its distribution different from that of other nodes but for its unexpected statistical information. Whether a node is an outlier or not is examined with centrality index. The larger the node has centrality indexes, the more probabilistically it may be identified as an outlier. The experimental results on artificial data sets demonstrated that two algorithms are very efficient and effective.

As an active microwave sensor, synthetic aperture radar (SAR) is capable of continuously monitoring geophysical parameters related to the structural and electrical properties of the earth's surface and subsurface. With the development of advanced SAR technologies with high resolutions and multiple imaging modes, SAR generates a large amount of remotely sensed data to be transmitted and processed. The raw data compression has become important tools to reduce the huge amount of data for downlink and required memory on-board. In this paper, four compression algorithms are discussed, including block adaptive quantization (BAQ) algorithm, amplitude and phase compression (AP) algorithm, wavelet BAQ (WT-BAQ) algorithm and wavelet packet BAQ (WPT-BAQ) algorithm. Considering the statistical independent property between amplitude and phase of raw data along with the growing popularity of wavelets, two additional algorithms are presented: wavelet AP (WT-AP) algorithm and wavelet packet AP (WPT-AP) algorithm. The six different algorithms are compared in the signal domain and the image domain with several quality parameters and the simulation is given to validate analytic result. The experimental results will be used for remotely sensed data acquisition, data processing and SAR systems design.

This paper is focused on the task of building extraction from high resolution imagery, which is primarily comprised of two steps. The first one is the building location by modified pose clustering, and the second is building extraction using a novel matrix search algorithm. As a generate-and-test algorithm, pose clustering produces some building hypotheses based on vote accumulation, aimed at image subsets likely to contain just a building. After building hypotheses verification, some false alarms could be eliminated based on geometric rules. Then we focus on image subsets, each of which is a potential region containing a building. Most buildings are comprised of orthogonal and sequential corners. We classify the corners into four types according to the orientation of corresponding edges. Each type of corners is labeled with a tag for identification, such as ABCD, etc. Building contained in each image subset can be represented as a tag sequence. Based on the tag sequence and the matrix formed by the dominate line sets, we develop an efficient matrix searching algorithm to address the task of extraction. The experiments carried out in our system show the promising potential of this scheme.

Resolving sub-pixel area information extraction has gained increasing attention in the remote sensing community. Automated morphological endmember extraction (AMEE) which integrates spatial and spectral information to select endmembers, is able to provide a relatively good characterization of general landscape conditions. As tradition support vector machine (SVM) predicts only class label, in order to obtain the abundance fractions of targets of interest, SVM method can be combined with pairwise coupling. This paper describes a model which combines SVM approach and AMEE algorithm. One of the main advantages of using this model is that it performs automatic sub-pixel information detection. At last, simulated and real Landsat TM data are used to demonstrate the potential of this approach.

Extracting target information from the remote sensing images has been becoming an important method of updating the spatial geography data. With the development of spatial technology, sensor technology, digital image processing technology and the computer pattern recognition technology, how to extract target information from the high resolution remote sensing images has become the target of many researchers. Based on the feasibility experimental study of road extraction using Mathematics Morphology, this paper put forward one kind of road extraction method with Mathematics Morphology as primarily path and seed growing as auxiliary path. City road network information in high resolution remote sensing image is taken as the research object. In this paper Mathematics Morphology method and the segment method of Support Vector Machine are used. This paper presents that the combination of Mathematics Morphology and seed growing method has priority to Mathematics Morphology or seed growing used respectively, especially has the superiority in extracting the road detail information.

In all ways of perception, vision is the main channel of getting environmental information for intelligent virtual agent (IVA). Reality and real-time computation of behavior simulation of intelligent objects in interactive virtual environment are required. This paper proposes a new method of getting environmental information. Firstly visual images are generated by setting a second view port in the location of viewpoint of IVA, and then the target location, distance, azimuth, and other basic geometric information and semantic information can be acquired based on the images. Experiments show that the method gives full play to the performance of computer graphic hardware with simple process and higher efficiency.

Taking a sub-area of semi-arid west Jilin Province as example, we mainly discuss the method of shelter forest extraction in sandy area from Landsat-7 ETM+ imagery in this study. After the comparison of the image fusion methods including HIS transforms, PCA transforms, Brovey transforms and Wavelet transforms, the method of Brovey transforms improved by wavelet analysis is presented for further processing. The details information in fused ETM+ image by this improved method is more considerable and fruitful. Using unsupervised classification in combination with supervised classification and threshold method based on NDVI, we extract the farmland shelterbelts from the fusion image finally. The accuracy of classification is more than 85%. From the experiment result, this method shows a better performance in the shelter forest extraction in a typical semi-arid sandy.

Information extraction from high-resolution remote sensing image automatically has been attracting wide audience globally nowadays. Traditional pixel-based classification for remote sensing image with high spatial resolution is out need of precision. Considering of the characteristic of remote sensing, object-oriented approach gives the resolution. Taking Quickbird image as an example, we extract some typical urban targets from the image using object-oriented image analysis in this study. The most suitable scale of image segmentation is also discussed. We also evaluate the classification precision in associated with different segmentation scale. Result shows that object-oriented approach has a great deal of advantage such as high precision, high efficiency, convenience and so on. When the segmentation scale is defined between 15 and 20, we will get the best classification result. Extracting at the most suitable scale of image segmentation, the precision of classification can reach above 90 percent.

Shadow in panchromatic band of some high spatial resolution remotely-sensed data can be worked at building height information extraction. This paper introduced an operational method to undertake height distribution information extraction using SPOT-5 panchromatic and multi-spectral data. Whether such a method is feasible and sensibilities of result to some influential factors are analyzed. Then, the method is applied to process SPOT-5 data of Beijing in year 2004 and 2007. Height distribution result shows that high-rise building within Chaoyang District contributes a lot to height growing of the whole city.

Since vector approach can be applied for accurate geo-processing, while raster approach is suitable for spatial analysis, the integration of raster and vector approaches has been studied for years. For spatial analysis, data mining or other geo-processing, it is often necessary to retrieve the entities in GIS databases frequently. However, due to lacking of the description of spatial relations among the entities in current studies, these retrievals are severely time-consuming. This paper is to promote an integrated approach for geographic feature retrieve in a mechanism called "raster: relation-vector: entity" method concerning both the process speed and information maintenance. Firstly, a "dimension-plus" relational raster is designed for keeping all the identity information of the original spatial object based on object-oriented data model. "dimension-plus" means one more dimension is employed to store more information. Then scanning technique is developed for detecting the relations of the spatial objects in this new raster. Topological information is observed in a foreseeable raster index time. Finally topological information is transferred to vector organization and complex geometric objects can be reconstructed using vector data with minimal time consumption. This research realizes the recognition and the rebuild of spatial entities that are described in spatial shape, layer identity and the individual characteristics (e.g. color and style) of each entity in the map of .dwg format, both of the geometric information and semantic information are kept well in the retrieve process.

Ellipse feature is widely accepted as one of the most fundamental features and its extraction in image has very important significance in many fields. We addressed the problem of extraction of ellipse with high accuracy in this paper. An ellipse extraction algorithm based on line diffusion function model has been proposed using local gray value variation to precisely identify edge location. This algorithm can automatically extract ellipse and carry through high-accurate ellipse localization. This paper firstly analyzed the primary principle of line diffusion function mode, then described the implementation of high-accurate ellipse feature extraction algorithm based on line diffusion function model. During the process, the mathematical model is provided and high-accurate ellipse feature extraction algorithm has been developed. Finally, the proposed approach is tested with real imagery and experimental results are presented to demonstrate the efficiency and accuracy of the proposed algorithm.

High Quality Prime Farmland (HQPF) is high, stable yields based on land consolidation of prime farmland, and has its important impact upon China's food security. To make clear the status-in-quo of the HQPF is important to its construction and management. However, it is difficult to get the spatial distribution information of the constructed HQPF enough rapidly in mountainous area using ground investigation, as well as hard to satisfy the requirements of large-scale promotion. A HQPF extraction framework based on object-oriented image analysis is discussed and applied to aerial imageries of Tonglu County. The approach can be divided into 3 steps: image segmentation, feature analysis & feature selection and extraction rules generation. In the image segmentation procedure, canny operator is used in edge detection, an edge growth algorithm is used to link discontinuous edge, and region labelling is carried out to generate image object. In the feature analysis & selection procedure, object-oriented feature analysis and feature selection methods are also discussed to construct a feature subset with fine divisibility for HQPF extraction. In the extraction rules generation procedure, the C4.5 algorithm is used to establish and trim the decision tree, then HQPF decision rules are generated from the decision tree. Compared with supervised classification (MLC classifier, ERDAS 8.7) and another object-oriented image analysis method (FNEA, e-Cognition4.0), the accuracy assessment shows that the extraction results by the object-oriented extraction patters have a high level of category consistency, size consistency and shape consistency.

This paper presents a new method applied to texture feature representation in RS image based on cloud model. Aiming at the fuzziness and randomness of RS image, we introduce the cloud theory into RS image processing in a creative way. The digital characteristics of clouds well integrate the fuzziness and randomness of linguistic terms in a unified way and map the quantitative and qualitative concepts. We adopt texture multi-dimensions cloud to accomplish vagueness and randomness handling of texture feature in RS image. The method has two steps: 1) Correlativity analyzing of texture statistical parameters in Grey Level Co-occurrence Matrix (GLCM) and parameters fuzzification. GLCM can be used to representing the texture feature in many aspects perfectly. According to the expressive force of texture statistical parameters and by Correlativity analyzing of texture statistical parameters, we can abstract a few texture statistical parameters that can best represent the texture feature. By the fuzziness algorithm, the texture statistical parameters can be mapped to fuzzy cloud space. 2) Texture multi-dimensions cloud model constructing. Based on the abstracted texture statistical parameters and fuzziness cloud space, texture multi-dimensions cloud model can be constructed in micro-windows of image. According to the membership of texture statistical parameters, we can achieve the samples of cloud-drop. By backward cloud generator, the digital characteristics of texture multi-dimensions cloud model can be achieved and the Mathematical Expected Hyper Surface(MEHS) of multi-dimensions cloud of micro-windows can be constructed. At last, the weighted sum of the 3 digital characteristics of micro-window cloud model was proposed and used in texture representing in RS image. The method we develop is demonstrated by applying it to texture representing in many RS images, various performance studies testify that the method is both efficient and effective. It enriches the cloud theory, and proposes a new idea for image texture representing and analyzing, especially RS image.

A fuzzy edge detection algorithm based on object-cloud and maximum fuzzy entropy principle are proposed in this paper. According to the uncertainty of the objects in the RS image, the spatial objects in RS image space can be mapped to the cloud space by 1:M cloud model. Object-cloud will have the digital characteristics to describe the fuzziness and randomicity of objects in RS image. According to the cloud operation, boundary-cloud and its digital characteristics can be achieved and the membership matrix of transition region can be constructed. By maximum fuzzy entropy principle, edge detection can be accomplished in the membership matrix of transition region.

In this paper, we propose an automatic and accurate image registration method for the high resolution. Due to the strong distortion caused by the terrain relief in this kind of images, it cannot be resolved by one set of transformation coefficients for the whole image. So the method mainly consists of two parts: one part is the dense feature point matching, and the other is the faced based differential registration. The matching algorithm integrates the feature point matching, relaxation optimization technique, the Least Square Matching, the coarse-to-fine strategy, and it can provide hundreds of thousands of reliable and accurate control points. With the TIN, these points divide the image into a lot of small triangles. For each triangle, we can assume the local distortion is simple and can be depicted by the affine transformation function. Finally, faced based differential registration is performed to resample the slave image. Experiments have been carried out and satisfactory results have been obtained.

In this article, the SIFT method is employed to combine a lot of images together from aerial unmanned airplane, without any control points. And the image number of this method is smaller than that of the triangulation. SIFT feature, which has shown great success in computer vision, is introduced into image registration in remote sensing. We extract distinctive invariant features from images that can be used to perform reliable matching between different views of an object or scene. The features are invariant to image scale and rotation, they are well localized in both the spatial and frequency domains, reducing the probability of disruption by occlusion, clutter, or noise.

Clouds in satellite images must be precisely identified prior to any further analysis in any case. A new cloud detection algorithm based on spectrum analysis and snake model is put forward in this paper. According to the distinction of the spectral bands for the MODIS and the spectrum curve of different objects, we can differentiate the cloud from other objects well by the band1, band6 and band26. Because the distinct difference between cloud and the earth's surface, the detected thresholds have good robustness, and they are not sensitivity to images. Then we can optimize the cloud boundary by snake model. We adequately use the image information of the three bands in snake model via color gradient. By balancing these model-based and data-driven energy terms using regularization parameters, the snake algorithm can extract very accurate cloud boundaries without gaps and spurious branches. According to numerous experimental results, the new cloud detection algorithm in this paper is simple, feasible and suitable.

Attitude estimation method is one of influencing factors for the attitude accuracy. Traditionally, the elements of the rotation matrix as attitude unknowns are estimated optimally, but the solved attitude angles based on the elements of rotation matrix aren't optimal. A rigorous attitude estimation approach for satellite attitude determination based on star sensor is presented in this paper, which directly considers three-axis attitude angles as attitude unknowns. The experiment indicates the proposed approach can improve the attitude accuracy to a great degree when the position errors of image points are within ±0.5 pixel, and the efficiency can be guaranteed as well.

There are kinds of methods for ortho-rectification in application of remote sensing, including Collinearity Equation Model, Strict Geometric Model based on Affine Transformation, Improved Polynomial Model, Rational Function Model, Method based on Neural Network, and so on. But there is lack of system comparison between these methods. On the basis of detailing the algorithm of these methods above, advantages and drawbacks about these algorithms are summarized in this paper. Specific emphasis is the mathematical derivation and algorithm of RFM. Two kinds of algorithm based on neural network were taken in application of ortho-rectification. To compare accuracy and effective between the above methods, we also detailed the processing steps and make some experiments. The result shows that: in the condition of the same GCPs distribution, Rational Function Model that can reach sub pixel accuracy is the best of all from the viewpoint of precision, which can be used in practice in spite of its relatively slower speed.

In the process of remote-sensed image fusion with wavelet packet transform, wavelet basis with different properties can exhibit different fusion performance. It is significant to find the best wavelet packet basis and apply it in the process. However, for image fusion, best basis searching algorithm must works within two wavelet packet trees, in the case that the present algorithm only works within one tree. The paper firstly proposes a new searching algorithm working in two trees, then realizes a new image fusion method using wavelet packet transform with the best basis that is developed from the new algorithm. Experiment testifies: under the fusion rule based on texture, the method develops more advantage of wavelet packet transform, and gains a better fusion performance compared with other image fusion method using wavelet packet transform (including wavelet transform).

The noise in natural images sometimes changes according to imaging mechanism or local image information. This is called spatially varying noise. It is obvious that classical variational denoising algorithms such as the Rudin-Osher-Fatemi model are not suitable for this kind of noise. We propose a variational method to remove this spatially varying noise based on the estimation of local variance for a given image, such that high noise regions are smoothed meanwhile the textures and certain details in low noise regions are preserved. Moreover, we give the proof of existence of the minimizer of our proposed functional. The experimental results show visual improvement and high signal-to-noise ratio over other variational denoising models.

Synthetic Aperture Radar (SAR) images have been widely used in remote sensing applications. However, the SAR image contains speckle noise. In this paper, we propose a variational method to delineate the boundary of objects in SAR images. It is implemented by two steps: The first step is speckle reduction and the second step is boundary detection. The two steps are implemented with a unified frame of energy minimization. In each step, we define an energy functional, and the corresponding minimizer of the functional is regarded as the result. An ERS-2 PRecision Image (PRI) over Proserpine area in Australia is used to demonstrate the algorithm. The results of both steps appear to be very promising.

The remote sensing technology has become the important information source in environment investigation, Moreover, optical satellite images are the most important information source. Although the optical satellite images may provides high resolution, multi-spectral images and better vision images than active satellite, the disadvantage is affected by the atmospheric condition easily. In general, the cloud cover is the most common noise, may decrease the image information abundantly and has impact on the environmental monitoring application seriously. According to the cloud imaging model, add defilade manually with different reflection coefficient to simulate different thickness of cloud. Then utilize GIS analytical method and cooperate with histogram calculation to extraction different reflection coefficients boundary. In this research, we get the upper threshold limitation value for haze and lower threshold limitation value for thick heavy cloud. So, we change the classification level from 2 ordinal levels into 3 qualitative levels. We change the thick and haze cover classification into threshold limitation value heavy, haze and fuzzy could cover classification by using the Formosat-2 satellite images. Make use of therefore way, can change the description yardstick into the quantitative yardstick that is we change the ordinal scale into interval scale in the image of cloud cover efficiency.

China has launched her second ocean color satellite HY-1B on 11 Apr., 2007, which carried two remote sensors. The Chinese Ocean Color and Temperature Scanner (COCTS) is the main sensor on HY-1B, and it has not only eight visible and near-infrared wavelength bands similar to the SeaWiFS, but also two more thermal infrared bands to measure the sea surface temperature. Therefore, COCTS has broad application potentiality, such as fishery resource protection and development, coastal monitoring and management and marine pollution monitoring. Atmospheric correction is the key of the quantitative ocean color remote sensing. In this paper, the operational atmospheric correction algorithm of HY-1B/COCTS has been developed. Firstly, based on the vector radiative transfer numerical model of coupled oceanatmosphere system- PCOART, the exact Rayleigh scattering look-up table (LUT), aerosol scattering LUT and atmosphere diffuse transmission LUT for HY-1B/COCTS have been generated. Secondly, using the generated LUTs, the exactly operational atmospheric correction algorithm for HY-1B/COCTS has been developed. The algorithm has been validated using the simulated spectral data generated by PCOART, and the result shows the error of the water-leaving reflectance retrieved by this algorithm is less than 0.0005, which meets the requirement of the exactly atmospheric correction of ocean color remote sensing. Finally, the algorithm has been applied to the HY-1B/COCTS remote sensing data, and the retrieved water-leaving radiances are consist with the Aqua/MODIS results, and the corresponding ocean color remote sensing products have been generated including the chlorophyll concentration and total suspended particle matter concentration.

Colored dissolved organic matter (CDOM) is one of the most important optical components which affect the sea surface spectral. Because of the similar spectral absorption properties between CDOM and organic detritus, it is difficult to separate them, especially in the coastal water with high turbidity. In many study, the total absorption coefficient of CDOM and organic detritus was combined as a optical parameter retrieved by the remote sensing algorithms. In this paper, a quasi-analytic remote sensing algorithm of CDOM has been developed, which can separate the absorption between CDOM and organic detritus. Firstly, the absorption spectrums of CDOM and organic detritus have been analyzed using the in-situ optical dataset measured by the Case II water optical investigation program in China Yellow Sea and East Sea in the spring, 2003. And then, a quasi-analytic remote sensing algorithm of CDOM has been developed to derive the absorption coefficients of the CDOM and organic detritus separately from the total absorption coefficient which can be retrieved from ocean color remote sensing data directly. The algorithm has been validated using the in-situ optical dataset of China Yellow Sea and East China Sea in the spring of 2003, and the synthesis optical data set and global in situ data set from IOCCG, and the results show that this algorithm perform well.

Texture and shape analysis offer interesting possibilities to characterize the structural heterogeneity of classes in the high spatial resolution satellite imagery. In this paper, texture features are generated based on the Gaussian Markov random field (GMRF) model, and shape features are measured using geometric moments. Then feature selection is implemented according to the class separability. To reduce the border blurring effect introduced by texture features, the unsupervised classification algorithm involved ordered procedures is proposed, in which linear objects are extracted using spectral and shape features firstly, then other objects are detected using the combination of spectral, texture, and shape features. The proposed classification method is implemented using QuickBird imagery. For comparison, the standard K-means method with spectral data is used as a benchmark. The experimental results show that the ordered classification method with the combination of spectral, texture, and shape information performed better than conventional methods.

As is well known, Interferometric synthetic aperture radar (InSAR) has been widely used in remote sensing field, which can reflect actual topographic trend or possible surface deformation. The precision of interferometric phase is critical to the final measurement. Due to the orbit attitude influence, such phase difference between the scattering elements on the same height level, which is named as flat-earth phase, usually causes the complex interferogram dense and difficult to be used in further procedures. Before phase unwrapping, interferogram must be flattened to derive accurate topographic or deformation information. Traditional methods pose problem to retrieve accurate flat-earth phase, which finally lead to inaccurate elevation or deformation information. A new algorithm of flat-earth phase removal is proposed in this paper based on SAR satellite system geometry and spectrum information of actual interferogram. The basic procedure of the method is firstly introduced and then the test results are following listed. From the comparison between the new algorithm and conventional ones, some advantages can be easily shown: The whole calculation can be easily understood and applied; accurate flat-earth phase can be retrieved and removed not only airborne SAR image but also satellite SAR image, which will improves the quality of complex interferogram to be well unwrapped.

Image analysis tools can be used to extract ground information from grey image which transformed from LiDAR point cloud. In this paper, a new algorithm based on interest value is presented to process the LiDAR data. According to this algorithm, the values of slope between each hit point pair within one grid are calculated. Then the feature point is chosen by interest value. At last, the grid elevation value is transformed into grey scale value. By utilizing the interest value, the optimal grid step can be obtained, and the principle components of point cloud slope information is reserved in grey image transformed by this algorithm. A set of LiDAR data of a tidal flat area, as a study area, are transformed into grey image by using interest value, and then edge detection and texture analysis are preformed to extract shoreline and low vegetation. Comparing with the results exported from other methods, experiments show that the algorithm based on interest value is better than the conventional methods. And it is possible that additional ground feature information such as texture information can be extracted from the interest grey image if appropriate image analysis tools are used.

Mapping surface sediment types is particularly challenging in muddy intertidal flat area due to muddy characteristics and tidal fluctuation. With the combination of Hyperion hyperspectral image and field survey data, two regression based image interpretation methods, namely characteristic band method (CBM) and band differential method (BDM), were used for sediment type classification and mapping. It was found that under low tidal level there was a strong correlation between surface sediment reflectance and its sand, silt and clay contents in shortwave infrared band. For 2102nm wavelength, the correlation coefficient by former method reached -0.8954, 0.9070 and 0.6547 respectively while the latter method had a relatively lower correlation capability. So choosing this band as the characteristic band, three linear regression models were constructed and the sand, silt and clay contents were quantitatively inversed from their corresponding reflectance values. A linear equilibrium corrective method was then applied to some "bad" pixels for inversed contents amendment due to regression model's linear transforming limitation. Based on these corrected component contents, Shepard triangular classification method was adopted and the sediment types for the whole intertidal flat were automatically obtained with a high interpretation precision of 87.9%. Results showed that the hyperspectral remote sensing reversion method could be well utilized for dynamic monitoring and analyzing of the depositional environment changes in muddy intertidal flat region.

In this paper, we investigate the performance of an endmember extraction algorithm when it is implemented in different fashions. The implementation fashion is changed by the use of a dimensionality reduction process, parallel or sequential mode. This results in four different versions of a single algorithm. We take the Automatic Target Generation Process (ATGP) algorithm as a study case due to its excellent performance. The experimental results show that a dimensionality reduction process can not only reduce computational complexity but also improve performance by compacting useful information into a low-dimensional space; the parallel mode can provide better performance than the sequential mode with the increase of computational complexity. Instructive recommendations in the selection or implementation of endmember extraction algorithms for practical applications are provided.

Vegetation abundance is an important indicator of urban heat island (UHI), because it influences the partitioning of sensible and latent heat fluxes. In order to reveal the effect of vegetation abundance on UHI of Wuhan city, one of the fast changing urban area in China, we classified land use/land cover types and calculated land brightness temperature (LBT) from a Landsat Enhanced Thematic Mapper Plus (ETM +) image acquired on July 9, 2002. The vegetation fraction derived from a linear spectral mixture analysis (LSMA) model was used as an alternative indicator of vegetation abundance. The fractal analysis of LBT and vegetation abundance was also conducted on 20 transects. Results showed that the spatial pattern of LBT changed with vegetation abundance and higher temperature was located in the area of lower vegetation abundance. Unmixed vegetation fraction was more negatively correlated with UHI than NDVI for most land cover types, except for water. Fractal analysis of image texture showed that transects comprised of larger number of different land cover types exhibited higher fractal dimension. On the contrary, the fractal dimension was lower in transects that covered mostly by built-up land. In addition, the fractal dimension correlation between LBT and vegetation abundance was higher than that between LBT and NDVI.

Because of the high water content of vegetation, water absorption feature dominate spectral reflectance of vegetation in the near-infrared region of the spectrum, and chlorophyll dominate the visible region. Previous studies have primarily related water band indices (WI) to vegetation water content. But the similar studies are vacancy in Rained Agriculture Areas of Loess. Two observation tests were carried out in arid and semi-arid area in Loess Plateau in order to find out the best preferential sensitively spectral index to water content and chlorophyll for the spring wheat and to monitor crops drought in this area. The results indicated that at leaf level the NDVI and EVI are the highest sensitive indices to the FMC and Chlorophyll, and for the leaf EWT, SAVI is the best index((r=0.738,P<0.01)); at canopy level, the red edge (λ_red) and the water content have the best relationship, and the sensitivity for WI₁₁₈₀ and NDWI are better. And the λ_red is also the best indictor for the chlorophyll at canopy level, the second is R670/R440, Furthmore, If considered the potential for atmospheric interference when data are collected from aircraft or satellite plarforms, So WI₁₁₈₀, WI₁₄₅₀ and NDWI may be the feasible for satellite remote sensing of vegetation water content at the canopy level. Meanwhile the NDVI and EVI may be the best index for satellite remote sensing of vegetation water content at leaf level for the arid and semiarid Rainfed Agriculture Areas of Loess Plateau.

With the advantage of image-spectrum integration and quantitative analysis, space-borne hyperspectral remote sensing technique was increasingly applied in ground object identification and information extraction at coastal region to solve the difficulty for field observation and sampling. In order to deeply excavate the embedded spectral information for different features in coastal area, the preprocessing process of hyperspectral image was essential and necessary. So taking Hyperion hyperspectral image as example dataset, the objective of this article is to study and build a doable flowchart for Hyperion image preprocessing to get the reflectance image of coastal region for further study and use. The processes include: (1) bad lines fixing; (2) vertical stripes removing; (3) atmospheric correction; (4) geometric correction and (5) tidal flat area separation from vegetation and water body. Related algorithms and parameters were also discussed in detail.

Hetao Irrigation District located in Inner Mongolia, is one of the three largest irrigated area in China. In the irrigational agriculture region, for the reasons that many efforts have been put on irrigation rather than on drainage, as a result much sedimentary salt that usually is solved in water has been deposited in surface soil. So there has arisen a problem in such irrigation district that soil salinity has become a chief fact which causes land degrading. Remote sensing technology is an efficiency way to map the salinity in regional scale. In the principle of remote sensing, soil spectrum is one of the most important indications which can be used to reflect the status of soil salinity. In the past decades, many efforts have been made to reveal the spectrum characteristics of the salinized soil, such as the traditional statistic regression method. But it also has been found that when the hyper-spectral reflectance data are considered, the traditional regression method can't be treat the large dimension data, because the hyper-spectral data usually have too higher spectral band number. In this paper, a partial least squares regression (PLSR) model was established based on the statistical analysis on the soil salinity and the reflectance of hyper-spectral. Dataset were collect through the field soil samples were collected in the region of Hetao irrigation from the end of July to the beginning of August. The independent validation using data which are not included in the calibration model reveals that the proposed model can predicate the main soil components such as the content of total ions(S%), PH with higher determination coefficients(R²) of 0.728 and 0.715 respectively. And the rate of prediction to deviation(RPD) of the above predicted value are larger than 1.6, which indicates that the calibrated PLSR model can be used as a tool to retrieve soil salinity with accurate results. When the PLSR model's regression coefficients were aggregated according to the wavelength of visual (blue, green, red) and near infrared bands of LandSat Thematic Mapper(TM) sensor, some significant response values were observed, which indicates that the proposed method in this paper can be used to analysis the remotely sensed data from the space-boarded platform.

Hyperspectral remote sensing image classification is a challenging task in remote sensing applications because this image always has some information redundancy and is easy to be affected by noise or lack of the separability. A semi-supervised classification method based on principal component analysis (PCA) method and kernel fuzzy C-means (KFCM) algorithm for hyperspectral remote sensing image is proposed in this paper. First the PCA method finds an effective representation of spectral signature in a reduced dimensional feature space. Then a semi-supervised kernel-based FCM algorithm, called SSKFCM algorithm by introducing semi-supervised learning technique and the kernel trick simultaneously into conventional fuzzy C-means algorithm, is introduced to classify the feature vectors. Finally numerical experiments are conducted on a hyperspectral remote sensing image that provides digital images of 80 spectral bands with wavelength rang from 455 nm to 1642 nm. Classification performance is estimated by classification accuracy and kappa coefficient. The simulation results show that the proposed approach can be effectively applied to hyperspectral remote sensing image classification.

The high-dimensional feature vectors of hyper spectral data often impose a high computational cost as well as the risk of "over fitting" when classification is performed. Therefore it is necessary to reduce the dimensionality through ways like feature selection. Currently, there are two kinds of feature selection methods: filter methods and wrapper methods. The former kind requires no feedback from classifiers and estimates the classification performance indirectly. The latter kind evaluates the "goodness" of selected feature subset directly based on the classification accuracy. Many experimental results have proved that the wrapper methods can yield better performance, although they have the disadvantage of high computational cost. In this paper, we present a Genetic Algorithm (GA) based wrapper method for classification of hyper spectral data using Support Vector Machine (SVM), a state-of-art classifier that has found success in a variety of areas. The genetic algorithm (GA), which seeks to solve optimization problems using the methods of evolution, specifically survival of the fittest, was used to optimize both the feature subset, i.e. band subset, of hyper spectral data and SVM kernel parameters simultaneously. A special strategy was adopted to reduce computation cost caused by the high-dimensional feature vectors of hyper spectral data when the feature subset part of chromosome was designed. The GA-SVM method was realized using the ENVI/IDL language, and was then tested by applying to a HYPERION hyper spectral image. Comparison of the optimized results and the un-optimized results showed that the GA-SVM method could significantly reduce the computation cost while improving the classification accuracy. The number of bands used for classification was reduced from 198 to 13, while the classification accuracy increased from 88.81% to 92.51%. The optimized values of the two SVM kernel parameters were 95.0297 and 0.2021, respectively, which were different from the default values as used in the ENVI software. In conclusion, the proposed wrapper feature selection method GA-SVM can optimize feature subsets and SVM kernel parameters at the same time, therefore can be applied in feature selection of the hyper spectral data.