Embedded Files
Dr. Kim's research on human motion classification by radar using machine learning is making impacts on academia, industry, and defense including (i) the non-cooperative classification of humans for security and sensing, such as intruder detection, senior fall detection, people counting, and patient monitoring, and (ii) cooperative classification for enhanced human-machine interface, such as hand gesture recognition, and speech recognition.
His recent research can be categorized to six, 1) Quantum computing for radar, 2) Development of deep learning algorithms for radar target detection/tracking/classification, 3) Human motion analysis using machine learning, 4) Bio-sensing using radar, and 5) Applied Electromagnetics
1) Quantum Computing for Radar
Y. Lee and Y. Kim, "Beampattern optimization of frequency diverse array using Quantum annealing," IEEE Transactions on Aerospace and Electronic Systems, Under review, 2025.
This study proposes optimizing the beampattern of Frequency Diverse Array (FDA) radar using Quantum Annealing (QA). We apply a QA approach to optimize the beampatterns of FDA by controlling the frequency offsets. Since QA only solves Quadratic Unconstrained Binary Optimization (QUBO) problems, the continuous frequency offset variables are binarized and transformed into QUBO form using a Factorization Machine (FM).
Y. Oh, Y. Lee and Y. Kim, "Auto-calibration of antenna array with Quantum annealing," IEEE Transactions on Aerospace and Electronic Systems, Under review, 2025.
This paper proposes a quantum annealing (QA)-based auto-calibration method that efficiently estimates calibration parameters of array antennas. In large-scale antenna arrays, auto-calibration is time-prohibitive due to the need to optimize an extremely large number of calibration parameters. The core approach of this paper is to transform the auto-calibration task into a quadratic unconstrained binary optimization (QUBO) model that can be solved by quantum annealing.
M. Seong and Y. Kim, "Non-orthogonal features for Quantum SVM-based robotic dog detection using micro-Doppler signals," IEEE Transactions on Aerospace and Electronic Systems, Under review, 2025.
This study proposes a method for effectively classifying quadrupedal walking robotic dog from other moving targets using micro-Doppler signatures. Building on the widely used Support Vector Machine (SVM), we apply a Quantum SVM (QSVM), which utilizes quantum circuit-based kernel computation. To efficiently utilize high dimensional micro-Doppler signatures for classification, we introduce a novel dimensionality reduction method that projecting data into non-orthogonal basis, which is expected to leverage the properties of quantum circuits
M. Seong and Y. Kim, "Image classification using CNN-QNN hybrid model with optimized correlated features," CVPR, Under review, 2025.
We propose optimizing the correlation among convolutional neural netowrk (CNN) features used as inputs to quantum neural networks (QNN) to enhance image classification accuracy. Unlike prior approaches that employ orthogonal decomposition as preprocessing, we intentionally introduce correlated features that are more physically compatible with QNN. This design leverages the QNN’s inherent ability to exploit quantum entanglement for representing correlated states—an advantage unavailable to classical neural networks.
2) Development of Deep Learning Algorithms for Radar
J. Choi, Y. Chun, S. Eom, D. Oh, and Y. Kim, "Enhanced radar false alarm mitigation in low-RCS target detection using time-varying trajectories on range-Doppler diagrams with DCNN," IEEE Transactions on Instrumentation and Measurement, Feb. 2025.
This research proposes detecting low-radar cross section (RCS) targets using the time-varying characteristics in the range–Doppler diagram with 3-D deep convolutional neural networks (3D-DCNNs), which significantly suppresses false alarms (FAs).
J. Cha, K. Yoo, D. Choi and Y. Kim, "Human presence detection using ultra short-range FMCW radar based on DCNN," IEEE Sensors Journal,
This study introduces a methodology for detecting human presence in close proximity using frequency-modulated continuous wave radar. We focus on discerning human presence against inanimate objects by analyzing target vibrations.
D. Lee, H. Park, T. Moon and Y. Kim, "Continual learning of micro-Doppler signature-based human activity classification," IEEE Geoscience and Remote Sensing Letters, Jan. 2021.
This paper investigated methods to update trained networks when new data or a new class is added. Through continual learning, the catastrophic forgetting problem can be overcome in micro-Doppler classification.
I. Alnujaim, D. Oh and Y. Kim, "Generative adversarial networks for classification of micro-Doppler signatures of human activity," IEEE Geoscience and Remote Sensing Letters, June 2019.
While applying ML to radar data, it is found that there is an intrinsic issue with radar data, that is data deficiency issue. This paper addressed this problem through data augmentation via generative adversarial networks (GANs).
J. Park, J. Rios, T. Moon and Y. Kim, "Micro-Doppler based classification of human aquatic activities via transfer learning of convolutional neural networks," Sensors, Nov. 2016.
This paper introduced the concept of transfer learning to improve radar data classification accuracy when the available data samples are limited. The pre-trained networks for optical images were successfully applied to radar images, leading to improve the classification accuracy.
Y. Kim and H. Ling, “Direction of arrival estimation of humans with a small sensor array using an artificial neural network,” Progress In Electromagnetics Research, vol. 27, pp.127-149, 2011.
This paper proposes an array processing algorithm based on artificial neural networks (ANNs) to estimate the directions of arrival (DOAs) of moving humans using a small sensor array. Due to the small number of array elements, the proposed technique was found to be effective for tracking targets when the number is limited.
3) Human Detection and Motion Classification Using Deep Learning
I. Choi and Y. Kim, "A Data-Efficient Graph Attention Network for Human Motion Sensing Using Micro-Doppler Spectrogram with Structural Representation," IEEE Sensors Letters, Dec. 2025.
This study proposes a Graph Attention Network (GAT) for human motion classification by analyzing a structure of spectrograms obtained from radar sensor rather than conventional deep learning algorithms such as CNN and RNN.
Y. Kim and T. Moon, "Human detection and activity classification based on micro-Doppler signatures using deep convolutional neural networks," IEEE Geoscience and Remote Sensing Letters, vol. 13, pp.2-8, Jan. 2016.
In this paper, he applied deep learning algorithms to human activity classification. He developed DCNN to process human micro-Doppler spectrograms to capture the significant features as well as to construct the decision boundary. This paper was featured as one of the most frequently accessed documents on the GRSS web for more than 3 years.
Y. Kim and B. Toomajian, "Hand gesture recognition using micro-Doppler signatures with convolutional neural network," IEEE Access, Oct. 2016.
This paper proposed the classification of human hand gestures measured by radar. Measure data was converted to spectrogram and the deep convolutional neural network was applied to classify the spectrograms. This paper became one of mile stone for the hand gesture recognition area. This paper is very actively cited in a short period.
Y. Kim, S. Ha and J. Kwon, "Human detection using Doppler radar based on physical characteristics of targets," IEEE Geoscience and Remote Sensing Letters, vol.12, pp. 289 – 293, Feb. 2015.
In this paper, the physical characteristic of a human was evaluated. On the basis of micro-Doppler period and range displacement, the stride of humans and animals was estimated. Those parameters were compared and the classification was conducted by SVM.
J. Rios and Y. Kim, "Application of linear predictive coding for human activity classification based on micro-Doppler signatures," IEEE Geoscience and Remote Sensing Letters, vol. 11, pp. 1831-1834, Oct. 2014.
This paper describes the application of a linear predictive code (LCP) to the time-domain radar signal to classify human activities. LCP allows a time-domain processing that does not requires transformation into a frequency domain. This approach could reduce the time cost in realizing real-time processing.
Y. Kim and H. Ling, "Human activity classification based on micro-Doppler signatures using a support vector machine ," IEEE Transactions on Geoscience and Remote Sensing , vol. 47, pp. 1328 –1337, May 2009.
This paper describes classifying human micro-Doppler signatures using SVM. Diverse human activities were measured using Doppler radar and significant features were extracted from micro-Doppler spectrogram. Then, SVM was designed to classify them. It became a milestone paper in this area.
4) Bio-Sensing Using Radar
C. Lee, C. Yoon, H. Kong, H. Kim and Y. Kim, "Heart rate tracking using Doppler radar with the reassigned joint time frequency transform," IEEE Antennas and Wireless Propagation Letters, vol. 10, pp. 1096-1099, Sep. 2011.
In this paper, he applied re-assigned joint time frequency analysis to the receive radar signal to identify heartbeat. The displacement of chest was measured by radar and the cardiopulmonary motion was estimated.
Y. Kim, "Detection of eye blinking using Doppler sensor with principal component analysis," IEEE Antennas and Wireless Propagation Letters, pp.1-4, Issue 99, Feb. 2015.
This paper proposed the measurement of eye blinking using radar. Using principal component analysis, the conscious and non-conscious blinking could be classified.
R. Khanna, D. Oh and Y. Kim, "Through-wall remote human voice recognition using Doppler radar with transfer learning," IEEE Sensors Journal, vol. 19, pp. 4571 – 4576, Feb. 2019.
This paper presents an application of radar to detection the vibration of vocal cord and movement of mouth to recognize the voice. Transfer learning has been used to improve the classification accuracy. Through-wall measurement has been also conducted. This paper was featured in the IEEE Innovation spotlight on Nov 2019.
Y. Kim and Y. Li, "Human activity classification with transmission and reflection coefficients of on-body antennas through deep convolutional neural networks," IEEE Transactions on Antennas and Propagation, Mar. 2017.
The channel between on-body antennas was used to detect the body movements. When the creeping wave is perturbed by body motion, the channel pattern has been analyzed by DCNN.
D. Bresnahan, Y. Li and Y. Kim, "Monitoring human head and neck-based motions from transmission coefficient of on-body antennas," IEEE Antennas and Wireless Propagation Letters, July 2018.
This paper describes the use creeping wave to sense the head and neck movement. The variation characteristics of creeping wave was captured and analyzed by DCNN.
B. Xu, Y. Li and Y. Kim, "Classification of finger movements based on reflection coefficient variations of a body-worn electrically small antenna," IEEE Antennas and Wireless Component Letters, pp. 1812-1815, Mar. 2017.
This paper discusses the classification of finger movements based on the reflection coefficient variations of a wrist-worn electrically small antenna. Dynamic time-warping (DTW) algorithm is employed to classify the measured reflection coefficient variations of the FCH for different finger activities.
5) Applied Electromagnetics
M. Nazaroff, G. Byun, H. Coo, J. Shin and Y. Kim, "2-D Direction of arrival estimation system using circular array with mutually coupled reference signal", IEEE Sensors Journal, vol. 18, pp. 9763-9769, Dec. 2018.
This paper discusses experimental results from an actual implementation of synchronizing a DOA circular array receiver system to mitigate DOA estimation error. To reduce phase errors at each channel, the reference signal is mutually coupled to antenna array elements.
Y. Kim, Y. Kim and S. Lee, “Linearized mixer using predistortion technique,” IEEE Microwave and Wireless Components Letters, vol. 12, pp. 204 -205, June 2002.
This paper proposes a predistortion technique to reduce Intermodulation Distortion (IMD) that occurs during the transformation process of the mixer. The unique feature of the present work is that down converted IF signals are linearized with the predistorted RF signals in the mixer.
Y. Kim and H. Ling, “On the optimal sampling strategy for model-based parameter estimation using rational functions,” IEEE Transactions on Antennas and Propagation, vol. 54, pp. 762-765, Feb. 2006.
This paper investigates several sampling strategies to achieve the least number of samples required for a given RMS error when using the MBPE for a fast frequency sweep. They are: i) uniform-like sampling; ii) random sampling; and iii) sampling based on the Fisher information theory. These three methods are evaluated against a reference optimal sampling solution.
Y. Kim, S. Keely, J. Ghosh and H. Ling, “Application of artificial neural networks to broadband antenna design based on a parametric frequency model,” IEEE Transactions on Antennas and Propagation, vol. 55, pp. 669-674, Mar. 2007.
This paper proposed an artificial neural network (ANN) to predict the input impedance of a broadband antenna as a function of its geometric parameters. Compared with performance of direct approach, ANN approach led to a tenfold reduction in the number of required EM simulations and was still able to maintain an acceptable level of accuracy.
Page updated
Google Sites
Report abuse