Deep Learning Makes Measuring Glucose in Food Easy

Machine Learning

Deep Learning Makes Measuring Glucose in Food Easy

Recently, deep learning technology, a branch of machine learning that uses artificial neural networks, has been attracting attention. A joint research team has developed a deep learning-based glucose detection method that is robust to changes in sample position. The results of the research were published in the journal. Lasers and Photonics Review.

Metamaterials are man-made materials with unique electromagnetic properties not found in nature that allow them to manipulate electromagnetic waves such as light and microwaves. A structure commonly used in designing metamaterials is the split ring resonator (SRR), which features a ring with a cleft in the middle. This design allows SRRs to absorb, transmit, or reflect electromagnetic fields of specific frequencies, amplifying the signal by interrupting the smooth flow of current and inducing electromagnetic resonance within the ring. SRRs are widely used in sensors, but their effectiveness is limited by inconsistent and unreliable measurements that are affected by factors such as temperature, humidity, and sample location.

In this study, the team aimed to address the problem of variations in the electrical signal of SRR-based sensors caused by changes in the position of the sample. They first optimized the sensor to amplify electrical signals in the frequency range of 0.5 to 18 GHz using a photolithography process, in which light is used to pattern semiconductors. The researchers then used deep learning techniques to enable the glucose sensor to learn from the electrical signals measured at different locations.

Based on this foundation, the research team developed a one-dimensional convolutional neural network (1D CNN) and used it to conduct experiments, demonstrating that the model can effectively compensate for errors caused by variations in sample position, achieving a mean absolute error (MAE) of 0.695% and a mean squared error (MSE) of 0.876%.

In experiments predicting the sugar content of actual fruit juices, including pineapple, Jeju citrus fruits, and Shine Muscat, a glucose measurement sensor using the research team's 1D CNN model demonstrated high accuracy with MAE of 0.45% and MSE of 0.305%. By overcoming the challenges inherent to SRR, the researchers developed a reliable glucose measurement sensor suitable for practical use.

Schematic showing glucose concentration sensing based on deep learning. Image courtesy of POSTECH.

“We succeeded in controlling the electrical signals that are sensitive to changes in the sample's position, improving the consistency and reliability of our glucose measuring device,” said Professor Roh Jun-sook of Pohang University of Science and Technology (POSTECH), one of the researchers.

“It is also noteworthy that this technology can be commercialized and mass-produced using photolithography processes already widely adopted in the semiconductor industry,” he said.

Top image credit:

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *