Based On Least Square Support Vector Machine Airborne Gamma Ray Energy Spectrum Segmentation Denoising Method Research

In the airborne gamma ray spectrum measurement, it is necessary to filter the airborne gamma energy spectrum to obtain the useful information of the gamma ray spectrum of the ground surface. To ensure that the original features of the energy spectrum can be eliminated as much as possible under the interference of the radioactive statistical fluctuation, the electronic noise and so on. Airborne gamma ray spectrum has the characteristics of low count rate and high randomness. To find a good denoising effect and adaptability of the method has become a problem of airborne gamma spectrum denoising. The traditional energy spectrum denoising method has better effect on some specific gamma spectrum, but it has its own limitation. Statistical methods can be used to obtain the intrinsic relations and laws of the characteristic peak spectral line and the noise spectrum through training, so as to achieve the purpose of reasonable denoising. Support vector machine(SVM) is a new generation machine learning method, which can solve the practical problems such as small sample, nonlinear and local minimum points. Least squares support vector machine(LS-SVM) is a form of support vector machines, compared with the standard support vector machine has a small amount of computation, the algorithm is simple and so on.In this paper, the domestic and foreign airborne gamma energy spectrum measurement technology, spectrum to denoising method and statistical learning theory, support vector machine algorithm are investigated and analyzed. On this basis, carried out based on least squares support vector machines of the airborne gamma energy spectrum segmentation denoising method, mainly research study contents and conclusions are as follows:1. Analysis of the traditional spectrum denoising method(analysis of gravity method, least squares smoothing method, Fu Liye transform, wavelet) features:(1) the center of gravity method, least squares polynomial smoothing method on mobile spectrum filtering, filtering wide spectrum, compressed the lower, broadening more serious. Increase the weight peak probability;(2) compared with other filtering methods, the use of ideal lowpass filter, Fourier transform filtering, peak edge count value is relatively large deviation;(3) selecting wavelet analysis method depends on the threshold, soft threshold noise filtering method fully, the original shape and hard threshold method can be preserved very well signal noise filtering, but not sufficient.2. Emphasizing the the selection of kernel function, and a modified kernel function is proposed. The experimental results show that the modified kernel function of LS-SVM model is better than other kernel function LS-SVM model in denoising ability.3.The two important parameters of LS-SVM model of optimization process, using 10 fold cross validation method to obtain the optimal kernel parameters- disciplinary factor C and kernel function width sigma. With the establishment of the learning ability and the generalization ability of support vector regression machine.4. Based on least squares support vector regression model, according to window way site distribution of airborne gamma energy spectrum is segmented regression fitting noise reduction and for segmentation using a weighted superposition cohesive to solve the the algorithm in full spectrum analysis on the defects.A large number of experimental data show that the method can significantly reduce the statistical fluctuation of airborne gamma ray spectrum, and has a good adaptability and generalization of the LS-SVM denoising method.