大学毕业论文---the-remote-sensing-image-fusion-method论文.doc

The Remote Sensing Image Fusion Method

ABSTRACT

Remote Sensing Image Fusion is one of the key techniques in the Remote Sensing (RS) domain. With the rapid development of the RSinformation fusion has been playing an increasingly important role. After a brief introduction to the RS Technology and RS information fusion, this paper describes the multi-spectrum image fusion in detail, with the emphasis on the PCA Fusion, the HIS Fusion and the Wavelet Fusion approach, whose mathematical foundation, principle and traits are explored in turn. Finally in the traditional image fusion is proposed on the basis of a new fusion method based on PCA and HIS, such as the new image fusion method fusion, and through experiment verification analyzed the new method is feasible.

KEY WORDS Remote Sensing；Remote Sensing mage Fusion；K-LFusion；HIS Fusion

1 Introduction
Information Representation different level, multi-sensor remote sensing image fusion can be divided into the pixel level fusion, feature fusion and decision levelfusion.Pixel level fusion is the process of comprehensive information directly to obtain the various pieces of remote sensing image pixels, so the image segmentation, feature extraction work on the basis of more accurate and better visual effect.Pixellevelfusion is intuitive, simple to operate, and the application of the most widely used. Pixel-level fusion method IHS transform, wavelet transform, principal component analysis (PCA) and Brovey transform method. Feature-level fusion is the imagefeatureextraction,edge, shape, contour, texture, and other information will be extracted to a comprehensive analysis of the fusion process. Decision-level fusion is the integration of a high level, often application-oriented decision support services.
2 The pixel-level image fusion
    Pixel level fusion process can be divided into four general steps: pretreatment, transform and inverse transform (reconstructed image). Papers assume that most of the studies pixel level fusion the fused image registration, but there are a number of research papers specializing in registration processing the transformation stage using the main square: PCA, sometimes also known as the PCT; HIS transform; multiresolution method pyramid algorithms and multi-resolution wavelet transform. Integrated phases will be fused to the integrated processing of the result of the transformation of the image, to obtain the final fused image. Integrated approach can be divided into: the selection method. I.e., according to certain rules, respectively, to select the transform coefficient with the image is fused to form a new set of transform coefficients; weighting method. I.e. some weighted average algorithm to the transform coefficients of the different fusion image consolidated into a new set of transform coefficients; optimization method. That is different depending on the application, to construct a performance evaluation of fusion effect, and the consolidated results of the performance optimal. Inverse transform stage is based on a transform coefficient obtained by the synthesis stage of the inverse transform operation, to obtain a fusing image.
3 Feature-level fusion
The feature fusion between pixel-level fusion and decision level fusion intermediate-level fusion. Feature-level fusion is based on the pixel level fusion, using the parameter template, statistical analysis, mode associated geometry associated target recognition, feature extraction, fusion method to exclude false characteristics to facilitate system judgment. Feature-level fusion in two studies, integration of the goal state data fusion and target characteristics. The layer fusion advantage achieved considerable information compression, to facilitate real-time processing, to maximize its fusion results given feature information needed by the decision analysis. Research methods commence from the cluster analysis, Dempster-Shafer reasoning method, Bayesian estimation method, neural network method. For multi-source images of the same surface features target feature images were extracted from the edge area, spectral, texture feature information; thus association between the image features as well as the location and description of the fusion features form a feature vector, which is more accurate land reflect the essential characteristics of the target, and to improve the remote sensing image classification and target description accuracy.
窗体顶端
Feature-level image fusion method:
A,Dempster-shafer reasoning method: DS method of reasoning structure,bottom-up divided into three levels:
The first level is the update (updated) information to be combined with a full time independent of a set of reports from the same sensor, the sensor in order to reduce the random error.
The second level is inferred logic sensor reports a certain credibility in some credible target report.
The third level is the synthesis, the synthesis of reports from several independent sensor for a total output of cluster analysis: mainly used for target identification and classification.
B,Bayesian estimation method: Bayesian inference given a priori likelihood estimation and additional evidence of conditions. Can update a hypothetical likelihood function. However, the method requires prior knowledge, and when the number of solvable assumptions and conditions related, it is very complicated.
C, entropy method: a new technology as a fusion the Contact information content of the measure, calculated with the assumption.
Feature-level target state data fusion is mainly used in the field of multi-sensor target tracking. Fusion System first preprocessing on the sensor data in order to complete the data calibration, then mainly achieved related to the parameter estimates of the state vector.
The feature level fusion advantages: to achieve a considerable information compression, is conducive to real-time processing, and because the characteristics are provided directly with the decision analysis, and thus can maximize fusion results given feature information needed by the decision analysis.
Feature-level image fusion Disadvantages: poor than pixel level fusion precision. Most C3I systems, data fusion research are expanded in the level.
4 The decision-making level image fusion
Decision-level fusion is the integration of a high level, it first for each data attribute specification, then the results of the fusion, the fusion property description of the target or the environment. The results provide the basis for command and control decision-making. Therefore, the decision-making level fusion must proceed from the needs of the specific decision-making problems, and take full advantage of the measurement object extracted feature fusion of feature information using appropriate fusion technology to achieve. Decision level fusion is the final result of three fusion is true then pecificdecision-making objectives, the fusion results directly affect the decision-making level.
5 The image Fusion Method Comparison
As the research object, different purposes, image fusion method can also arievd, its main steps are summarized as follows:
(1) Pretreatment: The selected features should be the same for the two types of image acquisition processing such data, unified data format, select registration feature point resolution. Denoising of a sequence, and enhance the tomographic image, created according to the target characteristics that the transformed error of the two data sets of a mathematical model of the minimum criterion;
(2) The data fusion database used: two-dimensional or three-dimensional case, the target object or area of ​​interest is divided. The features should be selected to achieve a certain error criterion corresponding to the two points on the image, which may be labeled the physical mark, it can be anatomical feature point;
(3) feature points for image registration: regarded as a three-dimensional reconstruction and display the number of data sets between linear or nonlinear transformation that transformed two data sets qualitative and quantitative analysis of the same physical marker error reaches the minimum ;
(4) integration of image creation: after registration of the two modes in the same coordinate system images will each useful information fusion expression into a mathematical model;
(5) Extraction: extract from the fusion image and measurement parameters to obtain the respective determination result.
Feature fusion has the advantage of fusion results thus achieved considerable information compression to facilitate real-time processing. Since the extracted features and decision analysis directly related to information compression to achieve a considerable decision analysis gives the characteristics of the information required. Most current data fusion systems are expanded on that level.
Decision level fusion has the advantage of image fusion technology through various characteristics of input information and the results described in the decision-making, so the decision level fusion small amount of data, anti-interference ability. The main advantage of decision level fusion can be summarized as follows:
(a) Communication and transmission requirements low, it is determined less by the data.
(b) High fault tolerance. Both robustness and accuracy of the integration, through appropriate fusion method be eliminated.
(c) Data requirements are lower. Can effectively reflect the full range of the target can be homogeneous or heterogeneous, the dependence of the sensor and requirements reduced.
(d) Analysis capability is very strong. Research data fusion architecture and environmental information to meet different application needs.
Because of the pre-processing and feature extraction have higher requirements, so the decision level fusion is costly.
We should remember that the purpose of fusion is to get higher resolution image, making the image more abundant spectral information, so we should be more practical applications utilize three image fusion method, dogmatism can not commit an error. Each other, interspersed with three fusion methods applied so that we applied research in remote sensing image fusion efficiency optimized to maximize efficiency.
6 Summary and Outlook
Remote sensing image fusion is an important tool in remote sensing image analysis. Multi-scale remote sensing image fusion by multiscale information complementary to eliminate redundancy and contradictions, can improve the accuracy and reliability of remote sensing information extraction to improve data utilization. Can improve the spatial resolution of the remote sensing image by the image fusion technique to enhance the characteristics of the target, and to improve the classification accuracy and dynamic monitoring and information complementary capabilities.
With the improvement of remote sensors, level of development and mathematical computer algorithm update, remote sensing image fusion will the Earth environmental monitoring, land use classification, and the development and utilization of the earth's resources to provide more new ways.