現今的磁力感測技術廣泛應用於許多不同的領域，特別是在磁力異常檢測。本研究利用低成本、體積小的微機電系統（MEMS）磁力感測器，提出一種垂直堆疊的磁感測陣列架構，結合特殊的特徵提取，及機率神經網路（PNN），來強化磁力異常檢測的效能，並透過MIAT方法論來進行系統架構設計，採用軟體高階合成的方式整合至微控制器單元（MCU）。本實驗蒐集一組無磁性物質的資料進行PNN訓練，並使用兩組相同數量的有無磁性物質各一半的資料進行測試，實驗顯示，在預測第一組測試資料的最佳超參數及閥值，驗證在第二組測試資料依然能達到很好的精確性，證明設定超參數及閥值，在本系統中的有效性。在磁力異常探勘實驗中，從原始資料的蒐集，到PNN測試完成的磁力異常值，均採用三維分布圖與真實資料的方式來呈現，對單一感測器與磁感測陣列進行了詳細的比較。結果確認了磁感測陣列相較於單一感測器不僅提供了更高的分辨率，也具備更高的可靠性和精確度。;Magnetic sensing technology is extensively applied across various fields, especially in detecting magnetic anomalies. This study leverages low-cost, compact microelectromechanical systems (MEMS) magnetic sensors to propose a vertically stacked magnetic sensor array architecture. This setup combines unique feature extraction and a Probabilistic Neural Network (PNN) to enhance the detection capabilities of magnetic anomalies. The system design is guided by the MIAT methodology, incorporating software high-level synthesis into the microcontroller unit (MCU). In our experiments, data without magnetic materials were collected for training the PNN, and two sets of data, each consisting of equal parts magnetic and non-magnetic materials, were used for testing. The experiments demonstrate that the optimal hyperparameters and threshold settings determined from the first test dataset maintained high accuracy in the second test dataset, validating the effectiveness of these settings in our system. The exploration of magnetic anomalies, from the collection of raw data to the completion of PNN tests on magnetic anomaly values, is visualized using three-dimensional distribution plots and real data, providing a detailed comparison between individual sensors and the magnetic sensor array. The results confirm that the magnetic sensor array not only offers higher resolution but also ensures greater reliability and precision compared to individual sensors.
