Machine Learning Techniques for THz Imaging and Time-Domain Spectroscopy

Physics-assisted machine learning for THz time-domain spectroscopy: sensing leaf wetness

Signal processing techniques are of vital importance to bring THz spectroscopy to a maturity level to reach practical applications. In this work, we illustrate the use of machine learning techniques for THz time-domain spectroscopy assisted by domain knowledge based on light-matter interactions. We aim at the potential agriculture application to determine the amount of free water on plant leaves, so-called leaf wetness. This quantity is important for understanding and predicting plant diseases that need leaf wetness for disease development. The overall transmission of 12,000 distinct water droplet patterns on a plastized leaf was experimentally acquired using THz time-domain spectroscopy. We report on key insights of applying decision trees and convolutional neural networks to the data using physics-motivated choices. Eventually, we discuss the generalizability of these models to determine leaf wetness after testing them on cases with increasing deviations from the training set.

Recent Advances in THz Detection of Water

The frequency range of terahertz waves (THz waves) is between 0.1 and 10 THz and they have properties such as low energy, penetration, transients, and spectral fingerprints, which are especially sensitive to water. Terahertz, as a frontier technology, have great potential in interpreting the structure of water molecules and detecting biological water conditions, and the use of terahertz technology for water detection is currently frontier research, which is of great significance. Firstly, this paper introduces the theory of terahertz technology and summarizes the current terahertz systems used for water detection. Secondly, an overview of theoretical approaches, such as the relaxation model and effective medium theory related to water detection, the relationship between water molecular networks and terahertz spectra, and the research progress of the terahertz detection of water content and water distribution visualization, are elaborated. Finally, the challenge and outlook of applications related to the terahertz wave detection of water are discussed. The purpose of this paper is to explore the research domains on water and its related applications using terahertz technology, as well as provide a reference for innovative applications of terahertz technology in moisture detection.

Machine Learning Identification of Organic Compounds Using Visible Light

Identifying chemical compounds is essential in several areas of science and engineering. Laser-based techniques are promising for autonomous compound detection because the optical response of materials encodes enough electronic and vibrational information for remote chemical identification. This has been exploited using the fingerprint region of infrared absorption spectra, which involves a dense set of absorption peaks that are unique to individual molecules, thus facilitating chemical identification. However, optical identification using visible light has not been realized. Using decades of experimental refractive index data in the scientific literature of pure organic compounds and polymers over a broad range of frequencies from the ultraviolet to the far-infrared, we develop a machine learning classifier that can accurately identify organic species based on a single-wavelength dispersive measurement in the visible spectral region, away from absorption resonances. The optical classifier proposed here could be applied to autonomous material identification protocols and applications.

Non-destructive evaluation and fast conductivity calculation of various nanowire-based thin films with artificial neural network aided THz time-domain spectroscopy

Terahertz time-domain spectroscopy (THz-TDS) has been utilized extensively to characterize materials in a non-destructive way. However, when materials are characterized with THz-TDS, there are many extensive steps for analyzing the acquired terahertz signals to extract the material information. In this work, we present a significantly effective, steady, and rapid solution to obtain the conductivity of nanowire-based conducting thin films by utilizing the power of artificial intelligence (AI) techniques with THz-TDS to minimize the analyzing steps by training neural networks with time domain waveform as an input data instead of a frequency domain spectrum. For this purpose, Al-doped and undoped ZnO nanowires (NWs) on sapphire substrates and silver nanowires (AgNWs) on polyethylene terephthalate (PET) and polyimide (PI) substrates have been measured for dataset creation via THz-TDS. After training and testing a shallow neural network (SSN) and a deep neural network (DNN) to obtain the optimum model, we calculated conductivity in a conventional way, and the prediction based on our models matched successfully. This study revealed that users could determine a sample's conductivity without fast Fourier transform and conventional conductivity calculation steps within seconds after obtaining its THz-TDS waveform, demonstrating that AI techniques have great potential in terahertz technology.

Advances in terahertz technology for cancer detection applications

Currently, there is an increasing demand for the diagnostic techniques that provide functional and morphological information with early cancer detection capability. Novel modern medical imaging systems driven by the recent advancements in technology such as terahertz (THz) and infrared radiation-based imaging technologies which are complementary to conventional modalities are being developed, investigated, and validated. The THz cancer imaging techniques offer novel opportunities for label free, non-ionizing, non-invasive and early cancer detection. The observed image contrast in THz cancer imaging studies has been mostly attributed to higher refractive index, absorption coefficient and dielectric properties in cancer tissue than that in the normal tissue due the local increase of the water molecule content in tissue and increased blood supply to the cancer affected tissue. Additional image contrast parameters and cancer biomarkers that have been reported to contribute to THz image contrast include cell structural changes, molecular density, interactions between agents (e.g., contrast agents and embedding agents) and biological tissue as well as tissue substances like proteins, fiber and fat etc. In this paper, we have presented a systematic and comprehensive review of the advancements in the technological development of THz technology for cancer imaging applications. Initially, the fundamentals principles and techniques for THz radiation generation and detection, imaging and spectroscopy are introduced. Further, the application of THz imaging for detection of various cancers tissues are presented, with more focus on the in vivo imaging of skin cancer. The data processing techniques for THz data are briefly discussed. Also, we identify the advantages and existing challenges in THz based cancer detection and report the performance improvement techniques. The recent advancements towards THz systems which are optimized and miniaturized are also reported. Finally, the integration of THz systems with artificial intelligent (AI), internet of things (IoT), cloud computing, big data analytics, robotics etc. for more sophisticated systems is proposed. This will facilitate the large-scale clinical applications of THz for smart and connected next generation healthcare systems and provide a roadmap for future research.

Pretreatment to terahertz absorption curves by narrow undulation constraint and Its quick implementation suggested by convex hull

In this work, a method of pretreating THz absorption curve is proposed, which leads to minimal range in absorption, reserving necessary undulation of curve for identification by convolutional neural network. The kernel thought of proposed method is about confining the undulation of curve with a pair of narrow parallel lines and solving their optimal position by consecutively rotation of normalized curve at two fixed points. A fast algorithm is further proposed based on features of convex hull, whose procedure is described in detail. The algorithm involves definition of some important point sets, calculating and comparing slopes and determining best choice out of 4 potential rotations. The rationality of searching critical point is illustrated in a geometric way. Additionally, the adaption of the method is discussed and real examples are given to show the capacity of method to extract nonlinear information of a curve. The study suggests that methods regarding computer graphics also contributes to feature extraction with respect to THz curve and pattern recognition.

A General Neural Network Model for Complex Refractive Index Extraction of Low-Loss Materials in the Transmission-Mode THz-TDS

The complex refractive index for low-loss materials is conventionally extracted by either approximate analytical formula or numerical iterative algorithm (such as Nelder-Mead and Newton-Raphson) based on the transmission-mode terahertz time domain spectroscopy (THz-TDS). A novel 4-layer neural network model is proposed to obtain optical parameters of low-loss materials with high accuracy in a wide range of parameters (frequency and thickness). Three materials (TPX, z-cut crystal quartz and 6H SiC) with different dispersions and thicknesses are used to validate the robustness of the general model. Without problems of proper initial values and non-convergence, the neural network method shows even smaller errors than the iterative algorithm. Once trained and tested, the proposed method owns both high accuracy and wide generality, which will find application in the multi-class object detection and high-precision characterization of THz materials.

Terahertz spatio-temporal deep learning computed tomography

Terahertz computed tomography (THz CT) has drawn significant attention because of its unique capability to bring multi-dimensional object information from invisible to visible. However, current physics-model-based THz CT modalities present low data use efficiency on time-resolved THz signals and low model fusion extensibility, limiting their application fields' practical use. In this paper, we propose a supervised THz deep learning computed tomography (THz DL-CT) framework based on time-domain information. THz DL-CT restores superior THz tomographic images of 3D objects by extracting features from spatio-temporal THz signals without any prior material information. Compared with conventional and machine learning based methods, THz DL-CT delivers at least 50.2%, and 52.6% superior in root mean square error (RMSE) and structural similarity index (SSIM), respectively. Additionally, we have experimentally demonstrated that the pretrained THz DL-CT model can generalize to reconstruct multi-material systems with no prerequisite information. THz CT through the DL data fusion approach provides a new pathway for non-invasive functional imaging in object investigation.

THz-ATR Spectroscopy Integrated with Species Recognition Based on Multi-Classifier Voting for Automated Clinical Microbial Identification

The demand for rapid and accurate identification of microorganisms is growing due to considerable importance in all areas related to public health and safety. Here, we demonstrate a rapid and label-free strategy for the identification of microorganisms by integrating terahertz-attenuated total reflection (THz-ATR) spectroscopy with an automated recognition method based on multi-classifier voting. Our results show that 13 standard microbial strains can be classified into three different groups of microorganisms (Gram-positive bacteria, Gram-negative bacteria, and fungi) by THz-ATR spectroscopy. To detect clinical microbial strains with better differentiation that accounts for their greater sample heterogeneity, an automated recognition algorithm is proposed based on multi-classifier voting. It uses three types of machine learning classifiers to identify five different groups of clinical microbial strains. The results demonstrate that common microorganisms, once time-consuming to distinguish by traditional microbial identification methods, can be rapidly and accurately recognized using THz-ATR spectra in minutes. The proposed automatic recognition method is optimized by a spectroscopic feature selection algorithm designed to identify the optimal diagnostic indicator, and the combination of different machine learning classifiers with a voting scheme. The total diagnostic accuracy reaches 80.77% (as high as 99.6% for Enterococcus faecalis) for 1123 isolates from clinical samples of sputum, blood, urine, and feces. This strategy demonstrates that THz spectroscopy integrated with an automatic recognition method based on multi-classifier voting significantly improves the accuracy of spectral analysis, thereby presenting a new method for true label-free identification of clinical microorganisms with high efficiency.

Component spectra extraction and quantitative analysis for preservative mixtures by combining terahertz spectroscopy and machine learning

Preservatives are universally used in synergistic combination to enhance antimicrobial effect. Identify compositions and quantify components of preservatives are crucial steps in quality monitoring to guarantee merchandise safety. In the work, three most common preservatives, sorbic acid, potassium sorbate and sodium benzoate, are deliberately mixed in pairs with different mass ratios, which aresupposedto be the "unknown" multicomponent systems and measured by terahertz (THz) time-domain spectroscopy. Subsequently, three major challenges have been accomplished by machine learning methods in this work. The singular value decomposition (SVD) effectively obtains the number of components in mixed preservatives. Then, the component spectra are successfully extracted by non-negative matrix factorization (NMF) and self-modeling mixture analysis (SMMA), which match well with the measured THz spectra of pure reagents. Moreover, the support vector machine for regression (SVR) designed an underlying model to the target components and simultaneously identify contents of each individual component in validation mixtures with decision coefficient R² = 0.989. By taking advantages of the fingerprint-based THz technique and machine learning methods, our approach has been demonstrated the great potential to be served as a useful strategy for detecting preservative mixtures in practical applications.

Deep neural network classification of in vivo burn injuries with different etiologies using terahertz time-domain spectral imaging

Thermal injuries can occur due to direct exposure to hot objects or liquids, flames, electricity, solar energy and several other sources. If the resulting injury is a deep partial thickness burn, the accuracy of a physician's clinical assessment is as low as 50-76% in determining the healing outcome. In this study, we show that the Terahertz Portable Handheld Spectral Reflection (THz-PHASR) Scanner combined with a deep neural network classification algorithm can accurately differentiate between partial-, deep partial-, and full-thickness burns 1-hour post injury, regardless of the etiology, scanner geometry, or THz spectroscopy sampling method (ROC-AUC = 91%, 88%, and 86%, respectively). The neural network diagnostic method simplifies the classification process by directly using the pre-processed THz spectra and removing the need for any hyperspectral feature extraction. Our results show that deep learning methods based on THz time-domain spectroscopy (THz-TDS) measurements can be used to guide clinical treatment plans based on objective and accurate classification of burn injuries.

Improving signal-to-noise ratio of a terahertz signal using a WaveNet-based neural network

When acquiring a terahertz signal from a time-domain spectroscopy system, the signal is degraded by measurement noise and the information embedded in the signal is distorted. For high-performing terahertz applications, this study proposes a method for enhancing such a noise-degraded terahertz signal using machine learning that is applied to the raw signal after acquisition. The proposed method learns a function that maps the degraded signal to the clean signal using a WaveNet-based neural network that performs multiple layers of dilated convolutions. It also includes learnable pre- and post-processing modules that automatically transform the time domain where the enhancement process operates. When training the neural network, a data augmentation scheme is adopted to tackle the issue of insufficient training data. The comparative evaluation confirms that the proposed method outperforms other baseline neural networks in terms of signal-to-noise ratio. The proposed method also performs significantly better than the averaging of multiple signals, thereby facilitating the procurement of an enhanced signal without increasing the measurement time.

Non-destructive Detection of Insect Foreign Bodies in Finishing Tea Product Based on Terahertz Spectrum and Image

Non-destructive testing of low-density and organic foreign bodies is the main challenge for food safety control. Terahertz time-domain spectroscopy (THz-TDS) and imaging technologies were applied to explore the feasibility of detection for insect foreign bodies in the finishing tea products. THz-TDS of tea leaves and foreign bodies of insects demonstrated significant differences in terms of time domain and frequency signals in the range of 0.3–1.0 THz. These signals were corrected by the use of adaptive iteratively reweighted penalized least squares (AirPLS), asymmetric least squares (AsLS), and baseline estimation and de-noising using sparsity (BEADS) for reducing baseline drift and enhancing effective spectral information. The K-nearest neighbor (KNN) and partial least squares discrimination analysis (PLS-DA) models showed the best performance after AirPLS correction with the prediction accuracy of 98 and 100%, respectively. In addition, the locations and outlines of insect bodies could be clearly presented via the THz-TDS image. These results suggested that THz-TDS spectroscopy and imaging provide an alternative tool for the detection of insect foreign bodies in finishing tea products.

Fast FMCW Terahertz Imaging for In-Process Defect Detection in Press Sleeves for the Paper Industry and Image Evaluation with a Machine Learning Approach

We present a rotational terahertz imaging system for inline nondestructive testing (NDT) of press sleeves for the paper industry during fabrication. Press sleeves often consist of polyurethane (PU) which is deposited by rotational molding on metal barrels and its outer surface mechanically processed in several milling steps afterwards. Due to a stabilizing polyester fiber mesh inlay, small defects can form on the sleeve's backside already during the initial molding, however, they cannot be visually inspected until the whole production processes is completed. We have developed a fast-scanning frequenc-modulated continuous wave (FMCW) terahertz imaging system, which can be integrated into the manufacturing process to yield high resolution images of the press sleeves and therefore can help to visualize hidden structural defects at an early stage of fabrication. This can save valuable time and resources during the production process. Our terahertz system can record images at 0.3 and 0.5 THz and we achieve data acquisition rates of at least 20 kHz, exploiting the fast rotational speed of the barrels during production to yield sub-millimeter image resolution. The potential of automated defect recognition by a simple machine learning approach for anomaly detection is also demonstrated and discussed.

Terahertz Imaging for Breast Cancer Detection

Terahertz (THz) imaging has the potential to detect breast tumors during breast-conserving surgery accurately. Over the past decade, many research groups have extensively studied THz imaging and spectroscopy techniques for identifying breast tumors. This manuscript presents the recent development of THz imaging techniques for breast cancer detection. The dielectric properties of breast tissues in the THz range, THz imaging and spectroscopy systems, THz radiation sources, and THz breast imaging studies are discussed. In addition, numerous chemometrics methods applied to improve THz image resolution and data collection processing are summarized. Finally, challenges and future research directions of THz breast imaging are presented.

Roadmap of Terahertz Imaging 2021

In this roadmap article, we have focused on the most recent advances in terahertz (THz) imaging with particular attention paid to the optimization and miniaturization of the THz imaging systems. Such systems entail enhanced functionality, reduced power consumption, and increased convenience, thus being geared toward the implementation of THz imaging systems in real operational conditions. The article will touch upon the advanced solid-state-based THz imaging systems, including room temperature THz sensors and arrays, as well as their on-chip integration with diffractive THz optical components. We will cover the current-state of compact room temperature THz emission sources, both optolectronic and electrically driven; particular emphasis is attributed to the beam-forming role in THz imaging, THz holography and spatial filtering, THz nano-imaging, and computational imaging. A number of advanced THz techniques, such as light-field THz imaging, homodyne spectroscopy, and phase sensitive spectrometry, THz modulated continuous wave imaging, room temperature THz frequency combs, and passive THz imaging, as well as the use of artificial intelligence in THz data processing and optics development, will be reviewed. This roadmap presents a structured snapshot of current advances in THz imaging as of 2021 and provides an opinion on contemporary scientific and technological challenges in this field, as well as extrapolations of possible further evolution in THz imaging.

Wind-Induced Pressure Prediction on Tall Buildings Using Generative Adversarial Imputation Network

Wind tunnel testing techniques are the main research tools for evaluating the wind loadings of buildings. They are significant in designing structurally safe and comfortable buildings. The wind tunnel pressure measurement technique using pressure sensors is significant for assessing the cladding pressures of buildings. However, some pressure sensors usually fail and cause loss of data, which are difficult to restore. In the literature, numerous techniques are implemented for imputing the single instance data values and data imputation for multiple instantaneous time intervals with accurate predictions needs to be addressed. Thus, the data imputation capacity of machine learning models is used to predict the missing wind pressure data for tall buildings in this study. A generative adversarial imputation network (GAIN) is proposed to predict the pressure coefficients at various instantaneous time intervals on tall buildings. The proposed model is validated by comparing the performance of GAIN with that of the K-nearest neighbor and multiple imputations by chained equation models. The experimental results show that the GAIN model provides the best fit, achieving more accurate predictions with the minimum average variance and minimum average standard deviation. The average mean-squared error for all four sides of the building was the minimum (0.016), and the average R-squared error was the maximum (0.961). The proposed model can ensure the health and prolonged existence of a structure based on wind environment.