svm

With the implementation of"high resolution earth observation" major project in China, The richgeographic information datas obtained by high resolution images provide moreservices for social development and human progress. Because of huge amount ofinformation conveyed by image, invest expensive costs and long time workloadsare ineluctable to manual segmentation of images for statistical data. In orderto provide convenience for subsequent application of information on landresources, automatic classification of high resolution images by computer isconsidered the better choice compared with manual segmentation. It is more conduciveto the extraction and recognition of surface information of earth. Comparedwith the traditional classification algorithm, the advantages for highresolution image classification are provided by the nonlinear classificationfunction of SVM (Support Vector Machines) algorithm. This kind method can beapplied to the high resolution image classification of small samples with fastclassification speed and high accuracy, and solves the current situation ofvisual interpretation and manual segmentation of images. Therefore, it isimportant to design and imply automatic classification system of highresolution images based on SVM, and making the SVM algorithm be engineering andpractical is important for automatic classification of high resolution to befast and autonomous.

The papermainly includes two parts: algorithm research and system implementation. Thealgorithm research is carried out around SVM theory, and the systemimplementation is completed in the VS2010 environment using C# programminglanguage. In the algorithm section, the mathematical principle of SVM algorithmand the principle of multi-class classification are studied. The purpose ofresearch is to obtain the spatial features which are more conducive to theclassification of high resolution images. TheSVM classifier model is designed and simulated for classification usingdifferent eigenvalues, and the classification results are visualized. The importance of simple color texture in high resolution imageclassification is verified. In the system part, after understanding the systemrequirements, the designing process of the classification system is cleared,and the system design scheme is constituted, including the overall design idea,the interface design and the module division. The algorithm design andengineering implementation of three important modules are completed, includingfeature extraction, sample training and data prediction. Finally, compared with other traditional classification algorithms, theperformance of SVM algorithm in high resolution image classification isverified. The results show that the SVM algorithm combining with the highresolution image data characteristics make full use of the advantage of SVMalgorithm in the case of small sample classification, including high accuracy,fast calculation speed and strong generalization ability. Therefore, automaticclassification system of high resolution images is characterized withexpandability, rapidity, practicability and autonomy, which provides animportant reference value for technology research and software development ofhigh resolution image classification.

 

Key words  High resolution image;  Automatic classification system;  SVM;  Sampletraining;  Feature extraction