Evaluating the two-stage adaptive multi-threshold pulse extraction method for loose particle detection from multiple perspectives

Existing loose particle detection research focuses on the training of high-performance classifiers and the creation of high-quality data sets in the downstream link, while neglecting the preprocessing of loose particle signals in the source link. In fact, stable extraction of sufficient and reliable pulses from loose particle signals is crucial, but few studies focus on pulse extraction methods. In this paper, the authors proposed a two-stage adaptive multi-threshold pulse extraction method for loose particle detection, which solves the problems in existing pulse extraction methods, including inconsistent threshold reference objects, unstable threshold reference objects, unclear threshold setting rules, ignoring pulses with small amplitudes, unstable pulse extraction, low reliability of pulse extraction, and not considering the continuous multi-pulse problem. In addition, for the first time, the authors newly proposed an evaluation method to comprehensively evaluate the pulse extraction effect of pulse extraction methods from multiple perspectives, including the number of pulses from the mathematical perspective, normalized signal-to-noise ratio from the signal perspective, and classification accuracy from the machine learning perspective. Test results on sealed relays show that, from the mathematical perspective, the proposed method is qualified and can extract sufficient numbers of pulses. From the signal perspective, the average normalized signal-to-noise ratio of the signals processed by this method is the highest. From the machine learning perspective, multiple representative classifiers achieve the highest component classification accuracy and material identification accuracy on data sets created from pulses extracted by this method, and significantly higher than the current highest component and material identification accuracy.