Multi-Input Deep Learning Based FMCW Radar Signal Classification

Cha, Daewoong; Jeong, Sung Hoon; Yoo, Minwoo; Oh, Jiyong; Han, Dong Seog

doi:10.3390/electronics10101144

Cited by 18 publications

(15 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Instead of deep learning methods, choosing basic machine learning methods, including SVM, dramatically decreases the test accuracy results. In addition, in related deep learning studies, range-Doppler or spectrogram plots are considered image data, so CNN is used to classify human activity 24 , 25 . However, CNN layers cannot measure the relation between time steps.…”

Section: Resultsmentioning

confidence: 99%

“…In addition, in related deep learning studies, range-Doppler or spectrogram plots are considered image data, so CNN is used to classify human activity. 24,25 However, CNN layers cannot measure the relation between time steps. At this point, with the help of bidirectional LSTM, the input flow is in both directions that preserve the future and the past information.…”

Section: Ablation Studymentioning

confidence: 99%

See 1 more Smart Citation

Ground target classification using mmWave radar with bidirectional long short-term memory networks

Gunes

Ege

Morgül

2022

J. Appl. Rem. Sens.

View full text Add to dashboard Cite

.We propose an approach for millimeter wave radar targets classification based on the concatenated spectrogram and range-Doppler features. A publicly available dataset that contains raw single radar data for fast- and slow-walking people is utilized. First, the received channels are summed to obtain a higher signal-to-noise ratio. Next, range-Doppler and spectrogram plots are obtained after the two-dimensional fast Fourier transform of raw radar signals. Then the spectrogram dataset is normalized and augmented using a mixup algorithm and fed into the proposed encoder and decoder structures sequentially for the classification task. In addition to the bidirectional long short-term memory layer, dropout and batch normalization layers are added for regularization in the proposed network. Parameter optimization is done for each network for a certain number of parameters. Finally, the results show that with the proposed approach, a 0.961 mean F1 score is obtained, and it outperformed some state-of-the-art methods, such as convolutional neural networks (CNNs) and hybrid CNNs–long short-term memory networks. An ablation study of the proposed method is also presented.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Ablation Studymentioning

confidence: 99%

Ground target classification using mmWave radar with bidirectional long short-term memory networks

Gunes

Ege

Morgül

2022

J. Appl. Rem. Sens.

View full text Add to dashboard Cite

show abstract

“…Deep learning combined with videos has been seen in works on intelligent surveillance, human-computer interaction, and video retrieval. [32][33][34][35] Deep learning enables computers to extract the high-level features in video images automatically and quickly, thereby saving a substantial amount of time and manpower. Hence, deep learning is far superior to conventional machine learning, especially when the data set is complex.…”

Section: Training With a Large Amount Of Datamentioning

confidence: 99%

Deep learning‐based intraoperative video analysis for supporting surgery

Bai

2023

Concurrency and Computation

View full text Add to dashboard Cite

SummaryThe application of surgical tool detection and tracking and surgical phase recognition is research hotspots at present. Surgical tool detection and tracking provides accurate real‐time information to surgeons, helping them plan surgeries and get advance warnings so that procedures can be safer. Surgical phase recognition can help improve the surgery efficiency, facilitate surgeons to make correct decisions, and help evaluate surgeon skill. In this article, we described the advantages and disadvantages of applying deep learning to surgical procedures. Then, we analyzed the advantages of deep learning in visual analysis, summarized the algorithms typically used in surgical tool detection and tracking and surgical phase recognition, and reviewed the results of these algorithms. The main problems are disclosed and the development trend of deep learning‐based intraoperative video analysis for supporting surgery is briefly discussed.

show abstract

“…Different from the traditional pulse radar system that periodically transmits short pulses, this paper adopts a 77 GHz frequency-modulated continuous wave (FMCW) radar to measure the distance and speed of the target by continuously transmitting a frequency-modulated signal (Chirp signal) whose signal frequency increases linearly with time [15]. At the same time, the Short Range Radio (SRR) band in the 77 GHz frequency band can provide a scanning bandwidth up to 4 GHz, and the range resolution and accuracy are significantly improved, which is conducive to the capture of subtle human movements.…”

Section: Millimeter Wave Radar Echo Modelmentioning

confidence: 99%

Human Motion Recognition in Small Sample Scenarios Based on Gan and CNN Models

Zhong¹,

Li²

2022

PIER M

View full text Add to dashboard Cite

In the research of radar-based human motion classification and recognition, the traditional manual feature extraction is more complicated, and the echo data set is generally smaller. In view of this problem, a method of human motion recognition in small sample scenarios based on Generative Adversarial Network (GAN) and Convolutional Neural Network (CNN) models is proposed. First, a 77 GHz millimeter wave radar data acquisition system is built to obtain echo data. Secondly, the collected human motion echo data is preprocessed, the micro-Doppler features are extracted, and the range Doppler map (RDM) is used to project the velocity dimension and accumulate the two-dimensional micro-Doppler time-frequency map data set of the human motion frame by frame. Finally, the deep convolution generative adversarial network (DCGAN) is constructed to achieve data augmentation of the sample set, and the CNN is constructed to realize automatic feature extraction to complete the classification recognition of different human motions. Experimental studies have shown that the combination of GAN and CNN can achieve effective recognition of daily human motions, and the recognition accuracy can reach 96.5%. Compared with the manual feature extraction, the recognition accuracy of CNNs is improved by 7.3%. Compared with the original data set, the system recognition accuracy based on the sample augmentation data set is improved by 2.17%, which shows that the GAN has an excellent performance in human motion recognition in small sample scenarios.

show abstract

Multi-Input Deep Learning Based FMCW Radar Signal Classification

Cited by 18 publications

References 30 publications

Ground target classification using mmWave radar with bidirectional long short-term memory networks

Ground target classification using mmWave radar with bidirectional long short-term memory networks

Deep learning‐based intraoperative video analysis for supporting surgery

Human Motion Recognition in Small Sample Scenarios Based on Gan and CNN Models

Contact Info

Product

Resources

About