SVM Inversion Method For Time Domain Airborne Electromagnetic Data

Airborne transient electromagnetic method(ATEM)is a kind of aircraft geophysical method with aircraft as carrier and electromagnetic induction as exploration mechanism.As the method has a large depth of exploration,large area of exploration,to overcome the advantages of complex terrain,it is particularly suitable for China's geographical conditions.At present,the airborne transient electromagnetic method has been widely used in geological mapping,mineral resources exploration and environmental monitoring and other fields.Because the airborne transient electromagnetic data belongs to the wideband signal,it is easy to be affected by many kinds of noise,especially the aviation transient electromagnetic data quantity is big and there are many influence factor,makes it difficult to explain it.All in all,the study of airborne transient electromagnetic signal processing and interpretation is still the hotspot and difficulty in this area.In this paper,we try to introduce the idea of machine learning,use SVM to study a large number of geo-electric models and their electromagnetic response,and realize the inversion of ATEM data and improve the accuracy of inversion.In addition.In order to obtain high-quality ATEM data,we also study the de-noising method combined with principal component analysis and wavelet transform to improve the signal-to-noise ratio.Specifically,the main contents of this paper are as follows:(1)De-noising method combining principal component analysis and wavelet transform.This method extracts the principal components of the airborne transient electromagnetic data by principal component analysis,then analyzes the principal components by wavelet analysis,and reconstructs the electromagnetic data with the principal components to achieve the purpose of suppressing the noise.The experimental results of forward modeling data with noise added show that the algorithm has better de-noising ability and can improve the signal to noise ratio by 10-14 dB.The experimental results of field data show that this method can suppress the random noise and high frequency noise of airborne electromagnetic data in space domain and time domain.In addition,the method has the advantages of good stability and high computational efficiency,and can meet the requirement of field processing data.(2)Support vector machine inversion of airborne transient electromagnetic primitive data.In this paper,the electromagnetic response of the two-layer and three-layer geo-electric model is calculated by the airborne transient electromagnetic forward program as a set of sample data.The sample data set is divided into two subsets and one subset is trained as support vector machine inversion sample set,and the other as a set of test samples.Then we use the sample set of two layers of geo-electric model to analyze the optimal parameter combination of support vector machine to find the best combination of inversion parameters.Finally,we use this parameter combination to complete the support vector machine training and inversion of the airborne transient electromagnetic data of the two-layer and three-layer geo-electric model respectively.Taking the two-layer geo-electric model as an example,the average relative error of the inversion results is 8.06%,and the relative error of the depth is 11.56%.(3)Support vector machine inversion of principal component of airborne transient electromagnetic data.Because the correlation coefficient between the adjacent data of the electromagnetic data is up to 0.9,and the characteristic components of the aeromagnetic data are extracted by principal component analysis.The support vector machine is used to find the mapping relationship between the feature component and the geological model.In this paper,the electromagnetic response of the two-layer and three-layer geo-electric model is calculated by the airborne transient electromagnetic forward program.The principal component is obtained by principal component analysis and used as the sample data set.The sample data set is divided into two subsets,one subset is used as a set of training samples for support vector machine inversion,and the other is used as a set of test samples.The support vector machine inversion experiment is carried out on the principal component of the forward data of the two-layer and three-layer geo-electric model.It is found that the result of inversion is basically the same as that of the support vector machine based on the original data,and its efficiency compared to the former has been improved.