Magnetic object recognition with magnetic gradient tensor system heading-line surveys based on kernel extreme learning machine and sparrow search algorithm

Abstract

We found that single heading-line surveys from magnetic gradient tensor system (MGTS) can be used to realize pattern recognition of magnetic objects, such as shape and posture, which can greatly improve the target detection efficiency compared with the two-dimensional grid measurement. Abandoning complex mathematical process, we measure and learn several training routes in advance, and use kernel extreme learning machine (KELM) and sparrow search algorithm (SSA) to recognize the target. The magnetic gradient tensor and its derived variables are analyzed for the sensitivity of the magnetization direction, and two types of characteristic attributes suitable for the target posture and shape categories are summarized. Time-domain waveform feature extraction from continuously sampled signals helps build datasets with corresponding category labels. Principal component analysis (PCA) is used to reduce feature dimensionality and improve classifier efficiency. Both simulation and experiment dataset have achieved 100% accurate recognition of the target posture and shape categories.