Magnetic Anomaly Detection of Adjacent Parallel Pipelines Using Deep Learning Neural Networks

Magnetic anomaly detection is becoming increasingly prevalent for detecting and locating the buried pipelines. The detection performance is often hindered by adjacent pipeline, near-field ferromagnetic objects and random noises. In order to overcome these obstacles, a magnetic anomaly detection method based on deep learning neural networks (DLNN) is proposed to decouple and denoise the integrated detection to accurately extract the magnetic anomaly of single pipeline. The theoretical derivation of the vertical component of magnetic anomaly was executed based on Poisson's equation. Then the integrated detection was simulated by summing magnetic anomalies of parallel pipelines and metal sphere, as well as white Gaussian noise. The DLNN was constructed with improved optimization design, and trained using supervised learning method. The results show that the proposed method exhibits almost immune to random noises, the prediction accuracy approaches to 90% with signal to noise ratio (SNR) of 30 dB. Meanwhile, the predictive accuracy is still above 80% with interferences both from near-field ferromagnetic objects and random noises with SNR of 30 dB. The method becomes practically significant in the development of geomagnetic inspection instruments for the adjacent parallel pipelines.