The Texture Extraction And Classification Of The Seismic Images

Under the influence of the tectonic movement, strata may present the geological phenomena such as faults and cracks, leaving the marks of the geologic changes. In the view of graphics, these marks can be seen as textures. Since the geological structure is different, the texture density and direction are also not identical. In other words, the different texture region reflects different geological structure. The break in the direction and structure of the texture means the break of the geological structure, which provides us some information for searching oil and gas. Texture analysis is a process that using certain image processing technique to extract the texture characteristic parameters, so as to acquire the quantitative or qualitative description of the image. It is a kind of method which reflects the homogeneous phenomena of the images in the visual identity that not dependent on the image color and brightness changes. In this thesis, we use the method of texture analysis to highlight the break region of the texture in the seismic images to identify the effective reservoir.In this paper, we firstly introduce the concept of the texture analysis, and then emphatically expound the major extraction method of the texture feature. According to the experimental comparison with the actual data, we determine choose the gray level co-occurrence matrix to extract the features of seismic images. On the process of the image texture classification, we consider that the number of the known training samples is rather limited, and different classes may not be a linear relationship between them. So we choose the support vector machine to accomplish the classification. At last of my thesis, using the actual data of a part of the Jingbian gas field, the results show that we have adopted the method of the gray level co-occurrence matrix to extract the texture features and applied the support vector machine to make classification, which has some remarkable effect in the reservoir prediction. The research results obtained in the thesis are as follows: 1) The method of gray level co-occurrence matrix has good stability and the recognition of the extracted textures'characteristics is strong. In the calculation of the characteristic value of the point, we take the point as the center to open a window. Through many times of experiment, we find that the effect is the best when the window size is 7 to 7 and 9 to 9.2) Base on the co-occurrence matrix, we calculate the characteristics of contrast, homogeneity, energy, entropy and correlation, but the value of the correlation is very small and lack of changes. Obtaining poor image quality, we don't adopt the correlation in image classification. In addition, through comparing with the values of the features, we find that the areas where the value of entropy high as well as the homogeneity low in the image are unanimous corresponding to the well position. In the view of the texture interpretation, low homogeneity and high entropy mean the value of the image is random and changes quickly, which also mean the geological structure complex and may contain oil or gas resources.3) Because of the known sample dates are very limited, and the relationships are not often linear dependent between the reservoirs and the attributes,so we choose the support vector machine to classification. In the thesis, we predict the reservoir parameters in the unknown area and success in dividing the work area into two parts: the effective reservoir and the invalid reservoir. From the results of the classification, most of the wells can fall on the effective reservoir, which are coincided with the actual gas well position.