Classification of Analyte in Aqueous Solution via High-Frequency Transmission Spectroscopy

High-frequency transmission spectroscopy (HFTS) is the fusion of traditional radiofrequency and microwave characterization techniques with deep learning models to perform classification of materials across a wide range of application domains. It is especially applicable to the classification of chemical compounds found in aqueous solution. A potential application domain of HFTS is the rapid detection of contamination in the water supply. Each year, over 800,000 people are killed as a result of contaminants in the drinking water supply. This work presents a proof-of-concept application of HFTS to the classification of contaminants in a water sample. The “contaminants” used in this work are a selection of ten chloride salt solutions prepared at a concentration of 0.15μM. S<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">21</inf> scattering parameter measurements were taken via a commercial vector network analyzer on a frequency range of 50 kHz to 500MHz. From these scattering parameter measurements, the magnitude response curves are calculated and used in the analysis. Visual exploratory data analysis is performed on the magnitude response curves to visually confirm that there are distinct patterns in the response of the various solutes which could be detected automatically through the application of a deep learning model. A multiclass convolutional neural network, composed of a single convolution layer and a fully connected linear layer, was developed and trained using cloud-hosted GPUs which achieved a testing accuracy of 92% during multiclass classification.