Contribuții la implementarea unui sistem de procesare și analiză de semnal utilizat în spectroscopia de rezonanță nucleară cuadrupolară / Contributions to the implementation of a signal processing and analysis system for nuclear quadrupole resonance spectroscopy

In the context of the unprecedented rise in terrorism and organized crime, nuclear quadrupole resonance has emerged as a technique for scanning prohibited substances that can meet the high requirements of security checkpoints. This paper aims to present several hardware and software solutions to improve the detection of prohibited substances by the nuclear quadrupole resonance technique, such as noise reduction, the compensation of the factors that influence detection and increasing the detection accuracy using signal processing and analysis techniques. These solutions are implemented in an equipment that ensures the processing and analysis of the response signal, intended for detecting illicit substances in security checkpoints. The evaluation of the developed system shows that the implemented solutions increase the signal-to-noise ratio by 31 dB compared to a system that does not integrate them. The response signal’s detection is performed with a deep learning algorithm trained on a data set consisting of 20,000 acquisitions obtained in multiple scenarios. The algorithm achieves a detection accuracy of 99.9%, that surpasses existing solutions and allows its use in the operational environment. Two enhancement solutions are proposed and validated by using transfer learning to extend the algorithm’s applicability to other spectrometers and autorencoders for denoising. The use of black-box optimization techniques is proposed to identify the optimal excitation sequences and increases the signal-to-noise ratio by 8 dB. The detection system was presented, awarded and patented nationally and internationally.