The Determination of Buried Magnetic Material From Various Heights: A Neural Network Application

This article aims at determining the anomalies created by buried magnetic materials on the horizontal component of earth's magnetic field through the KMZ51 anisotropic magnetoresistive sensor (AMR) series and determining the upper surface image of the material through the sensor series located in different positions per soil by processing the sensor data obtained due to anomaly at different heights from the soil surface using a feedforward artificial neural network (ANN) trained with the Levenberg-Marquardt (LM) backpropagation algorithm. In our study, in this direction, first, a mechanical scanning system ensuring the 3-D movement of the sensors, a data capture module to process the data transmitted from the KMZ51 AMR sensors and to transmit to the computer, and a magnetic measurement system composed of a computer software package to evaluate and record the data transmitted to the computer were established. Afterward, the magnetic materials with known magnetic, chemical, and geometrical properties were buried in soil containing 28.5% magnetic particles and 4.1% natural moisture with the help of the measurement system and the voltage changes in AMR sensors resulting from the anomaly were transmitted into the computer through the 2-D movement of the platform. This process was repeated for five different heights. The voltage values obtained were converted into a data matrix; then, the undesired noises in data resulting from the magnetic character of soil were cleared by filtering through the median and the base. Such cleaned data were converted into black and white images with a threshold value calculation method for conversion into black and white in grayscale histograms developed by “Otsu”; and the upper surface image of the buried material was determined. Finally, a feedforward ANN using the LM backpropagation algorithm was trained with the data measured from 0-, 1.5-, 3-, 4.5-, and 6-cm heights from the magnetic material; and the voltage values measured from 3- and 6-cm heights from the magnetic material were provided to the ANN as inputs, respectively, and thus, the voltage values for the same material at 0-cm height were found. Thereby, the upper surface image of the material could be determined by the same way as that obtained from 0-cm height even though the sensor series is at a different position compared with the soil surface through the ANN developed. Although many magnetic materials were tested, the results of a cuboid and a cylindrical material were provided within the scope of our study article and the performance of the ANN was discussed. In addition, the success of the ANN in different soil mixtures was confirmed for these two materials.