Research On Characteristic Analysis And Detection Method Of Weak Magnetic Anomaly Signal

Magnetic detection is based on the principle that the induced magnetic field generated by magnetic target will change the original electromagnetic distribution.Ferromagnetic target also changes the original electromagnetic distribution in the geomagnetic field,which is called magnetic anomaly.Magnetic anomaly has significance in various magnetic detection applications because it contains important information of the tested target.The actual target signal is often submerged in different environmental magnetic noises,which makes it difficult to identify,so the main research in this paper is the detection of weak magnetic anomaly signal.Firstly,a model of the target magnetic anomaly signal is proposed to derive the scalar signal representation on the measured coordinate system.And the OBF of the magnetic dipole model of magnetic anomaly signals is introduced from the perspective of feature dimension reduction.Then,PCA and extraction are performed on different target signal pattern distributions.Next,the time-frequency distribution with the maximum sample discrimination degree is selected as the input of the subsequent algorithm by analyzing and comparing the time-frequency distribution of mixed signals and the pure noises.Secondly,the correlation matching detection algorithm is performed on the unknown signal and the simplified features to judge the target signal with a certain threshold.The geomagnetic noise processed in this paper is generally colored noise,so the whitening filtering process based on autoregressive model is adopted.Then,the OBF correlation detection is performed on the mixed signal and the pure magnetic noise.PCA and LDA greatly simplify the features from the perspective of sample spacing and class spacing respectively.The paper will compare the discrimination of the two methods between mixed signals and pure magnetic noises.Thirdly,the method of transforming the original problem into image classification problem is proposed,which converts the original signal into a two-dimensional image signal through time-frequency transformation,fits the data distribution in the actual measurement environment,generates positive and negative samples,and the magnetic anomaly target signal is detected by using the CNN method.Then,the key parameters and the training-testing process of the CNN model are interpreted for the specific applications of magnetic anomaly detection.Finally,the paper will discuss the experimental design of magnetic anomaly detection based on CNN,and analyze the impact of training sets containing measured signals from different sources on model training.Then the optimized network structure is designed by analyzing the receptive field of the time-frequency map,the time and space complexity.Then the sample structure is optimized based on the analysis of training process.Finally,the model and correlation detection method are measured,and the superiority of the CNN to the correlation algorithm is verified.The specific implementation of CNN-based magnetic anomaly detection is completed eventually.