Digital inspection approach of overlapped peaks due to high counting rates in neutron spectroscopy

This paper is concerned with digital inspection of normalized neutron pulses in the presence of high counting rate. An approach including several digital signal processing algorithms is implemented for such purpose. Initially, block diagram programming through a Matlab environment is conducted for the generation of normalized neutron pulses that degraded with white Gaussian noise (WGN). In the second stage, identification and classification of overlapped neutron pulses are demonstrated. Features reduction is applied to the input neutron events using several methods. A fuzzy K-nearest neighbor (fuzzy-KNN) classifier is applied for matching of extracted features. The inspected peaks confirm that fuzzy-KNN achieves 99% accuracy with applied extracted features. Accuracy of fuzzy-KNN classifier is compared with artificial neural network (ANN). The ANN demonstrates 100% accuracy with feature reduction on expenses of computational time. In third stage, handling of identified overlapped neutron pulses due to high counting rates is carried out. Five algorithms are implemented. The proposed algorithms are sensitivity, scale-space peak picking (SSPP), threshold peak, robustness peak detection (RPD) and crest detection with undecimated discrete wavelet transform (UDWT). The threshold peak besides crest detection with UDWT algorithms introduced less accurate recovery. Finally, resolution enhancement of visualized spectrum with special emphasis on complex spectrum is demonstrated. Five algorithms are built for narrowing the normalized neutron spectrum. These algorithms are smoothing sharpening, power law, optimum sharpening factors, power converted area and segmented sharpen algorithms. Resolution enhancement of normalized neutron spectrum by the power law algorithm is the best observable sharpening form. The accomplished high accuracy of the proposed approach for neutron spectrometry provides higher confidential nuclear security within industrial applications.

Automated summary

This paper introduces a digital inspection approach for normalized neutron pulses in the presence of high counting rate, utilizing signal processing, feature extraction, and classifier matching. The fuzzy-KNN classifier outperforms artificial neural network in terms of computational time and accuracy.