In this article, we report our work on the development of a non- invasive, rapid, robust, and high-fidelity technique that can be used to discriminate between genetic variants. Our study focused on terahertz (THz) spectroscopy and imaging to distinguish between genetic variants of the Allium genus rapidly and accurately. This was done by measuring the cellular water dynamics of the samples by measuring their evaporation profiles using Laser Feedback Interferometry (LFI) with THz Quantum Cascade Lasers (QCL). The evaporation profiles of the samples were then processed to create trajectories in the amplitude-phase domain, which correlated with cell age, cell type, and the amount of water bound to biomolecules. This technique can differentiate between the members of the Allium genus. The presence of outliers was also studied to determine the effectiveness of the technique for different samples and to negate external influence. This was done to discern the extent of influence of cell biomechanics and biochemistry between genetic variants. We found that within a genus, different species would have different degree of interaction between cellular water and cell biochemistry, which could be clearly mapped out using THz-QCL-based LFI. Based on our observations, we propose that this method could be appropriate for observing minute alterations in cellular water dynamics in real-time, and in the future, has the potential to be employed for rapid and effective genetic discrimination in agricultural and genome conservation applications.
Precision agriculture has evolved over the years to meet the growing demand for agricultural productivity with limited resources, and with it, the smart irrigation techniques are also gaining traction. Management of water is critical since it is one of the most significant components of the photosynthetic process and hence an indicator of crop health and yield. Due to the high sensitivity of Terahertz (THz) radiation towards the presence of water, in current work, the moisture content in leaves from a Capsicum annuum plant is approximated using THz time-domain imaging. To overcome the effect of approximations and limitations in theoretical models, this work aims to generalize the prediction of moisture content in plants by simulating drought stress in an un-watered plant through a detached leaf. This is achieved through analysis of time-lapsed THz imaging of several leaves by employing a machine learning approach. The THz images are processed to retrieve pixels corresponding only to the flat lamina without the veins because of their morphological differences. The process is repeated for all instances of images as the leaf dries up. For predicting the moisture content in the leaf, transmittance of the selected pixels at selected frequencies ranging from 0.4-2.1 THz are used to train supervised machine learning regression (SMLR) model. Standard error of estimate (SEE) used for performance analysis of Decision Tree, Random Forest and Support Vector regression models show that as the drought sets in and the leaves dry up, the prediction accuracy improves.
Paper, as a hygroscopic dielectric material, does not have specific spectral signatures in the Terahertz (THz) range from 0.2-6 THz. However, because of its constituent materials, including dry matter, moisture, and air pockets, it absorbs THz radiation, similar to biological tissues and green leaf. Though the absorption loss is not significant, varying levels of dampness in wet paper are observed over time using continuous wave (CW) based THz Spectroscopic system to quantify the moisture content of wet paper relative to paper at ambient environment. For this purpose, effective medium theory (EMT) approaches including Bruggeman (BM), Landau–Lifshitz–Looyenga (LLL), and Complex Refractive Index (CRI) models are analysed. However, EMT models are dependent on physical and optical properties of paper and water, which are not well-defined and are dependent on assumptions, approximations and rigorous calculations. To remove such dependencies, supervised machine learning regression (SMLR) algorithms in the form of decision tree (DT), random forest (RF), and support vector regression (SVR) are investigated. The conditioning of the training parameters is dependent on spectroscopic data which reduces the processing time and improves efficiency due to elimination of approximations. Prediction efficiency of SMLR models is observed to be better than that of EMT models. RF shows the best results in terms of coefficient of determination, 𝑅2 but the time required for training is more when compared to DT and SVR models. DT models show consistent performance, while predictions using different SVR models show variance with 𝑅2 ranging from 0.42 to 0.98.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.