The Texture Extraction and Classification of the Seismic Images
|School||Chengdu University of Technology|
|Course||Signal and Information Processing|
|Keywords||Feature Extraction Texture analysis GLCM Wavelet Transform Support Vector Machine (SVM)|
Strata affected by tectonic movement, generates geological phenomena such as faults, fractures, thus leaving the imprint of the geological history of changes. These traces from the graphics, that they are the texture. Of course, due to the different geological formations, the density of the texture is not the same direction, it can be said that the different texture regions reflect the different geological structure. In those texture direction or structural mutations in place, but also means the geological structure of the mutation, these information for finding oil or gas are important. Texture analysis refers to the use of certain image processing techniques to extract the texture parameters, to obtain a quantitative or qualitative description of the image process. It is not dependent on image color or brightness changes reflected in the image processing method of the visual characteristics of the homogenization phenomenon. In this paper, the use of texture analysis methods, to highlight seismic image texture mutations area, so as to achieve the purpose of identifying effective reservoir. This paper first introduces the basic concepts, then focuses on several major texture feature extraction methods, test comparison of the actual data, to determine seismic image feature extraction GLCM texture analysis. Image texture classification and identification, taking into account the known number of training samples is small and may not have a linear relationship between the different categories, so this paper support vector machine. Finally, the actual data for some sections of the Jingbian Gas Field, GLCM extracted texture features and application support vector machine classification, test results show that the study made in the work area of ??the reservoir prediction good results. Thesis achieved the following: 1) the GLCM method has good stability, extracted texture features are also very strong ability to identify the different blocks. In the calculation of the characteristic value of each point, to the point of the center fenestration. Through many trials contrast, found that when the size of the window, select 7 × 7 or 9 × 9 best results. 2) on the co-occurrence matrix, contrast, inverse difference moment, energy, entropy, and the correlation coefficient characteristic correlation coefficient taking into account the characteristics of the value is very small and the lack of change, the poor image effects, using no image classification. Furthermore, by comparing the distribution of the values ??of the various features can be seen when the entropy takes a high value, the the deficit moments low value, correspond exactly to the location of the wells. From the point of view of the meaning of the texture features, indicating where the value of random, local image changes quickly, which also indicates that the geological structure of the area is complex and may contains oil and gas resources. 3) due to the sample data is less known, and between the reservoir and the properties often do not show a linear relationship, this study using support vector machine reservoir parameters to predict the unknown region of the work area, and the success of the work divided into two parts the effective reservoir invalid reservoir. From the classification results, most of the well can fall within the effective reservoir, the predicted results with actual test wells bit more consistent.