Research On Interference Mitigation And Target Detection Of Airborne Magnetic Anomaly Detection

Airborne magnetic anomaly detection(MAD)is a technology using airborne magnetometer to detect the ferrous target which can distort the local ambient magnetic field.This technology originates from the 1930 s,and nowadays has been widely used in many fields,such as geological exploration,unexploded ordnance(UXO)detection,humanitarian aid,and archaeology.A complete airborne MAD system contains three parts,an aircraft,sensors,and a device used for data acquisition and post-processing.In this study,the focus is the post-processing technology,which is aimed at comprehensively processing the collected data and determining the existence of the magnetic anomaly signal.To realize this purpose,first,the interference terms should be removed as far as possible to enhance the signal quality,then the features of target signals should be detected to determine whether targets exist.Therefore,the post-processing technology of airborne MAD contains two main parts,interference mitigation and target detection.Because the magnetic field is omnipresent,the interference components are complex and hard to be modeled.Commonly,the aircraft magnetic field is handled separately,as it can be modeled.This technique is called as aeromagnetic compensation.Other interference components,which are mainly caused by the earth ambience,are usually mitigated through filtering.However,the model of aircraft magnetic field is collinear and time-varying,which makes the calculated magnetic field imprecise.Besides,the method of filtering the interferences outside the target signal frequency band has weak pertinency and limited effect.Additionally,because of the real-time requirement,most traditional target detection methods are based on signal processing,the detection rates are unsatisfactory.Based on the previous description,the main works of this dissertation are as follows:Firstly,in the aircraft magnetic model,the variables are strongly related with each other,this property,called as multicollinearity,interferes the accuracies of the calculated model coefficients and limits the compensation effect of aircraft magnetic field.This dissertation analyzes the sources of relations and finds that the variables causing the multicollinearity are different in different headings,based on which,this dissertation proposes a multimodel compensation algorithm.This algorithm builds a sub-model for a special heading through selecting the effective variables,weakens the multi-collinearity,and improves the compensation performance.Secondly,in the regression equation of calculating the aircraft magnetic model coefficients,the noise term is formed by the time-varying ambient magnetic field,which makes the noise term autocorrelated,interferes the coefficients accuracies,and leads to the deficient compensation performance.In comprehensive consideration of timevarying,this dissertation proposes a recursive aeromagnetic compensation algorithm.This algorithm is based on an adaptive compensation method,changes the mathematical type of the regression model,weakens the autocorrelation of the noise,keeps the leastsquare estimator effective,and lastly improves the compensation performance.Thirdly,to mitigate the ambient magnetic interferences outside the target magnetic band,filters with fixed bands are always applied but have deficient effects,as the target signal frequency varies with the flight parameters.This dissertation analyzes the expression of target signal and proposes an algorithm to adaptively calculate the narrower frequency band of the target signal,based on which,this dissertation designs a filtering algorithm through wavelet transform.This filtering algorithm can improve the performance of reducing the ambient magnetic field and can greatly increase the signal to noise ratio(SNR)of the filtered signal.Fourthly,the traditional target detection methods are commonly based on signal processing approaches and extract features point-by-point,however,this type conflicts with human intuition,makes the detection sensitive to noise,and leads to the unsatisfactory detection rate.From the perspective of pattern recognition,this dissertation reconstructs the representation of target signal and proposes a detection algorithm based on deep neural network(DNN)to extract the features of target signal.Compared with the traditional algorithms with manually designed features,this algorithm has stronger abilities on feature representation and target detection.