| Target recognition technology has been a hot research topic, which involves knowledge of image processing, data mining and pattern recognition, etc. So it is a strong cross-cutting technology. Target recognition technology is widely applied in military activity, industrial production and profoundly influences daily life. The main work of this paper was the application of relevant technology to identify the horizons in seismic image. Horizons can embody the shape of rock layers in seismic image, which play a fundamental role in forecasting and excavating energy and mineral underground. Automatic identification or auto-tracking horizon from massive seismic data is quite a challenging puzzle, of which the main reason is due to the lateral horizons of rapid change or fault affect make the continuity worse.Currently horizon tracking work is mainly done by manual. Defects in artificial interpretation are imprecise and heavily dependent on the interpreter’s subjective experience. Furthermore, artificial interpretation is low efficiency and can easily become a bottleneck restricting the entire interpretation work. In order to solve above problems, the development of the full horizon auto-tracking technology which have better performance gain more and more attention. This paper focused on characteristics of horizons, and proposed a new framework of two-dimensional full horizon auto-tracking, which is divided into two steps:1. Horizon Fragment Formation Algorithm based on spatial characteristics of horizons. Through a large number of statistic and observation on horizons, horizon distribution in space has the following characteristics: extrema belonging to the same horizon are the basic trend of continuous distribution and different horizons are roughly parallel in time direction because different geological horizons generate in different time. So density in same horizon is greater than that in different horizon. Based on above understanding, it translated horizon tracking issue to clustering problem based on density in space, so divided extrema with density high enough into the same horizon. Then horizon segments were formed. In order to make the clustering fragments comply with real horizons, this paper used mathematical morphology method to improve DBSCAN clustering algorithm.2. Horizon fusion algorithm based on horizon fragments. Clustering in step 1 only took into account the characteristics of the spatial continuity, so it cannot solve the problem that horizon fragments in the same horizon are separated in space. Due to the similar rock composition of same geological horizon, waveform characteristics of same horizon in seismic image are similar. According to the waveform reconstruction techniques, Chebyshev polynomial coefficients can be used to characterize features of the waveform. The waveform characteristics which are consistent with the hypothesis of Gaussian distribution can be clustered based on finite Gaussian mixture model which make fragments into complete horizons. Due to the complexity of the actual situation and the model errors, horizon segments cannot properly be classified to the corresponding horizons. This paper introduced the concept of segment confidence, then merging fragments process was flexible according to the size of confidence level.Finally, the proposed framework was applied to several practical seismic data, and the performance was evaluated by comparative analysis of auto-tracking method with artificial interpretation and commercial software. From the test results, the resolution of the method proposed by this paper is superior to the artificial interpretation ones. The overall effect has reached the level of commercial software. |
PDF Full Text Request