edfas.org 31 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 25 NO. 3 MULTILAYER PERCEPTRON DEVELOPMENT TO IDENTIFY PLASTICS USING FLUORESCENCE LIFETIME IMAGING MICROSCOPY Georgekutty Jose Maniyattu1, Eldho Geegy1, Maximilian Wohlschläger1, Nina Leiter1, Martin Versen1, and Christian Laforsch2 1Rosenheim Technical University of Applied Sciences 2University Bayreuth maximilian.wohlschlaeger@th-rosenheim.de EDFAAO (2023) 3:31-37 1537-0755/$19.00 ©ASM International® INTRODUCTION The way people live today is defined by plastics that offer alternative and long-term solutions to common societal needs and demands. More than 70% of the total plastics consumption in the European market is accounted for by the packaging, construction, and automotive sectors.[1] Due to the high demand, new plastic types having a higher resistance are developed every year. But as plastic manufacturers develop more resistible plastic types, the biodegradability is decreasing within a reasonable period of time, which harms nature and humanity.[2,3] Existing plastic analysis techniques like Fourier transform infrared (FTIR) spectroscopy and Raman spectroscopy are problematic because samples must be anhydrous. Also, FTIR and Raman spectroscopy can be affected by additives in the sample.[4] Identifying polymers utilizing their fluorescence lifetime is efficient. The fluorescence lifetime can be determined in the time-domain and in the frequency-domain.[5] Using frequency domain fluorescence lifetime imaging microscopy (FD-FLIM), identification of plastics is possible based on their characteristic fluorescence lifetime.[6] To automate the process of plastic identification an artificial neural network (ANN) is investigated. A databased classification using the measured data from the FD-FLIM camera is used for a multilayer perceptron (MLP). Different combinations of five channels obtained from the FD-FLIM camera are used to find the most useful features, train the neural networks, and evaluate the performance. THEORY FREQUENCY-DOMAIN FLUORESCENCE LIFETIME IMAGING Fluorescence lifetime imaging provides additional information on local physical and chemical parameters.[5] In FD-FLIM, the sample is excited using a harmonic sinu- soidal light resulting in a phase shifted re-emitted fluorescence signal measured at a known frequency ω.[5-7] Additionally, the fluorescence signal is damped in its amplitude and equivalent shifted. Comparing the amplitude and equivalent value of excitation and fluorescence signal, the modulation depth and modulation dependent fluorescence lifetime can be calculated. The FD-FLIM camera determines five images: fluorescence intensity (I), phase shift (P), modulation depth (M), phase dependent fluorescence lifetime (Pl) and modulation dependent fluorescence lifetime (Ml). The five channels and the metadata, which contain information about the camera used, the temperature of the camera at the time of acquisition and the exposure time, are stored in an ND2 file format, and each file has a size of 103.8 MB. ARTIFICIAL NEURAL NETWORKS ANNs consist of neuron layers, which can be described with an input layer, one or more hidden layers, and an output layer. The neurons are interconnected having an associated weight and threshold. A neuron is activated, passing the data to the next layer if the output of an individual neuron is above a specified threshold value.[8]
RkJQdWJsaXNoZXIy MTMyMzg5NA==