SRK 2022

POVZETEK
V tem prispevku predstavljamo avtomatsko klasifikacijo anorganskega pigmenta plastičnega materiala z uporabo terahertzčne spektroskopije in konvolucijskih nevronskih mrež. Plastične vzorce materiala postavimo med THz oddajnik in sprejemnik, pridobljeni THz signali pa se kategorizirajo s pomočjo nadzorovanega učenja. THz sistem generira frekvenčni odtis na frekvenčnem območju med 0,1 in 1,2 THz, ki je predstavljen z enodimenzionalnim (1D) vektorjem. Ta signal je skoraj nemogoče ovrednotiti s strojnim učenjem. Prispevek prispevek predlaga novo pred-obdelavo enodimenzionalnih THz podatkov, ki pretvori enodimenzionalne podatke v dvodimenzionalne, ti pa se učinkovito obdelajo z uporabo konvolucijske nevronske mreže. Predlagani algoritem za pred-obdelavo je sestavljen iz štirih korakov: odkrivanje vrhov, ekstrakcija ovojnice in postopek vzorčenja navzdol. Zadnji glavni korak uvaja okno s širitvijo spektra, ki preuredi enodimenzionalne podatke v dvodimenzionalne. S predlagano metodo preurejanja enodimenzionalnih podatkov v dvodimenzionalne je mogoče podatke obravnavati kot sliko. Tehnika širitve spektra zagotavlja robustnost klasifikatorja z zatiranjem pristranskosti meritev, zmanjšanje kompleksnosti nabora podatkov z zanemarljivo izgubo natančnosti in pospeševanjem klasifikacije z nevronsko mrežo. Eksperimentalni rezultati so pokazali, da je predlagani pristop dosegel visoko natančnost z uporabo klasifikatorja CNN in prekaša enodimenzionalno in dvodimenzionalno klasifikacijo THz podatkov z uporabo CNN, Support Vector Machine, Naive Bayesa in drugih metod.

ABSTRACT
This paper presents an automatic classification of plastic material’s inorganic pigment using Terahertz spectroscopy and convolutional neural networks. The plastic materials were placed between the THz transmitter and receiver, and the acquired THz signals were classified using a supervised learning approach. A THz frequency band between 0.1 and 1.2 THz produced one dimensional (1D) vector that is almost impossible to classify directly using supervised learning. This paper proposes a novel pre-processing of 1D THz data that transforms 1D data into 2D data, which are efficiently processed using a Convolutional Neural network. The proposed pre-processing algorithm consists of four steps: peak detection, envelope extraction, and down-sampling procedure. The last main step introduces the Windowing with Spectrum Dilatation that reorders 1D data into 2D data that can be considered as an image. The spectrum dilatation techniques ensure the classifier’s robustness by suppressing measurement bias, reducing the complexity of the THz dataset with negligible loss of accuracy, and speeding up the network classification. The experimental results showed that the proposed approach achieved high accuracy using a CNN classifier and outperforms 1D and 2D classification of THz data using CNNs, Support Vector Machine, Naive Bayes, and others.