Application In Geochemical Anomaly Identification With Blind Signal Extraction And SVM Method

Geochemical exploration is an important means of detecting mineral resources. Mathematical Geology treatment in regional geochemical exploration data is one of the important focuses of research. The primary task of geochemical exploration data processing is to determine the geochemical anomaly and its lower limits. In the tradition way of determination, it is taken as a prerequisite that the distribution law of geochemical elements satisfies with normal distribution or log-normal distribution. Then it takes the mean plus twice the variance as the lower limit. And this way gradually shows its limitation. As the earth covers an extremely complicated system, but the information in deep deposit is too limited, so it is necessary to continuously explore the mathematical geology treatment in geochemical anomaly.Blind signal processing is one of the hot spots of contemporary research and great success has been made in application of voice signal and image separation. The core issue of blind signal extraction is the learning algorithm about separation or solution of mixing matrix and it belongs to unsupervised machine learning. It draws out the statistically independent characteristics without the loss of information. Under the circumstance of want in priori knowledge, blind signal processing is a natural choice. From the perspective of algorithm, it is partial learning algorithm and comparatively simple. For the optimization criteria of Non-Gaussian metric, it is from the perspective of energy maximization to put forward the simultaneous blind extraction algorithm to extract simultaneously the signal of interest. When the signal source number is large, this method can greatly reduce the computation, obviously better than one by one extraction. So it is feasible to use blind signal extraction to identify the geochemical anomaly.The theoretical basis of Support Vector Machine is Statistical Learning Theory and basically, it does not include probability determination and the law of large numbers. Compared with neutral network, it has much superiority. First, the theoretical basis is mature. Moreover, based on the law of Structural Risk Minimization, it can sufficiently solve the problem of over-fitting in machine learning and has a feature of generalization. Theoretically, there definitely exist a global optimal solution and it will not fall into local optimum. Structurally, it employs the way of kernel function and it will not add the complexity when mapping to high dimensional space and effectively overcome the curse of dimensionality. And it is especially effective in small sample statistics without priori knowledge. For the processing of large sample statistics, there are many variations and improved algorithm to enable the processing of massive data.With careful study of blind signal separation theory and the characteristics of Support Vector Machine, based on the platform of MATLAB, this thesis combines the two factors , Firstly, with the simultaneous blind extraction algorithm to recognize the elements of variance then based on this, using support vector machines to separate the corresponding data and draws out the lower limit of anomalies, separates the anomaly data and delineates the abnormal concentration focuses. All these mark an innovative attempt in research of geochemical data processing. This thesis makes a comprehensive processing about the data in the tested region of 1:50,000 soils. From the real processing effect, the realization of this algorithm objectively and comprehensively reveals the actual mineralization situation. The Main elements of variance are consistent with the fact, and the lower limit of geochemical anomaly is reasonable, the delineated abnormal region basically coincides with the tested region. This shows that it is feasible to use Blind Signal Extraction algorithm and Support Vector Machine to the processing of geochemical data. The algorithm itself is effective and has referential and promotion value in research of processing way of geochemical anomaly identification.