Symmetric Magnetic Anomaly Objects' Orientation Recognition Based on Local Binary Pattern and Support Vector Machine

In order to identify the orientation or recognize the attitude of small symmetric magnetic anomaly objects at shallow depth, we propose a method of extracting local binary pattern (LBP) features from denoised magnetic anomaly signals and classifying symmetric magnetic objects that have different orientations based on support vector machine (SVM). First, nine component signals, such as magnetic gradient tensor matrix, total magnetic intensity (TMI), and so forth, are calculated from the original signal detected by the flux gate sensors. The nine component signals are processed by discrete wavelet transform (DWT), which aims to reduce noise and make the signal's features clear. Then we extract LBP texture features from the denoised nine component signals. From the simulation analysis, we can conclude that the LBP texture features of the nine component signals have good interclass discrimination and intraclass aggregation, which can be used for pattern recognition. Finally, the LBP texture features are constructed into feature vectors. The orientations of symmetric ferromagnetic objects underground are identified by SVM based on the feature vectors. Through experiments, we can conclude that the orientation recognition accuracy rate reaches 90%. This suggests that we can obtain the details of magnetic anomalies through our method.