Research On Denoising Of Potential Field Data Based On Deep Convolutional Neural Network

In the potential field survey,due to some inevitable factors,the measured field data usually contains obvious noise interference,which affects the subsequent data processing and interpretation.Especially in the complex environment,noise interference has the characteristics of high intensity and similar to the abnormal signal scale.Existing methods of denoising the potential field usually have a series of problems such as incomplete denoising or abnormal distortion of the bit field caused by the denoising transition,and it is difficult to obtain an ideal denoising effect.In response to this problem,an effective denoising algorithm needs to be developed.While removing noise interference,as much as possible to protect the effective anomaly signals,and independently find the best solution to achieve the purpose of improving the quality of the field data.As deep learning technology is widely used in the field of geophysics,this paper proposes a method of denoising data based on deep convolutional neural networks.The main research contents are as follows:We have deeply researched the domestic and foreign research status,basic theoretical content and technical principles of deep convolutional neural networks and potential field data denoising methods.On this basis,the most advanced denoising convolutional neural network Dn CNN in image denoising is applied to the denoising of potential field data.At the same time,a data set suitable for denoising the potential field data is established,and an improved algorithm that is more suitable for denoising the potential field data is proposed,that is,the mapping object is the abnormal signal of the potential field and the network depth of 15 layers.The experimental results show that,when dealing with strong noises with abnormal and similar noise scales,compared with the traditional denoising methods of potential field data,that is,experimenting with regularization filtering,smoothing compensation and sliding window averaging of different filtering parameters,the Dn CNN adaptive denoising algorithm constructed in this paper does not need to set the filtering parameters,and can not only protect the abnormal signal of the potential field,but also achieve the purpose of effectively removing strong noise globally.When analyzed from a quantitative perspective,the algorithm in this paper has better quantitative indicators(signal-to-noise ratio and mean square error).At the same time,the error comparison experiment also has a smaller error range.Finally,the applicability of the algorithm is tested.The test results show that the denoising algorithm in this paper has certain applicability to the noisy potential field data with different noise strengths and random abnormal signals.At the same time,comparing the denoising results under different strong and weak noise conditions,the algorithm in this paper is more suitable for denoising under strong noise conditions.This is of great significance for the subsequent processing and interpretation of the field data.