Tian-ao Cao

Dr. Tian-ao Cao

Harbin Institute of Technology
orcid

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
14 December 2020 in Measurement Science and Technology
Verified
Journal Article
Closed access
QWQisong Wang
Qisong Wang
Tian-ao CaoTian-ao Cao
Tian-ao Cao
DLDan Liu
Dan Liu
MZMeiyan Zhang
Meiyan Zhang
JLJingYang Lu
JingYang Lu
JSJinwei Sun
Jinwei Sun
In electroencephalography, multi-channel electroencephalogram (EEG) signals are usually utilized to improve classification accuracy. However, a large set of EEG channels increases the computational complexity, reduces the real-time performance and causes wearability difficulties. Channel selection methods have been widely investigated to reduce the number of channels with an acceptable loss of accuracy for EEG-based motor-imagery recognition. In this paper, we present a novel algorithm, called Support Vector Machine-Canonical Correlation Analysis-Channel Selection (SVM-CCA-CS). First, the energy features of the wavelet packet subnodes of the motor-imagery EEG signals are extracted. Then the weights of feature groups are calculated as initial channel weights, based on the CCA algorithm. The initial channel weights are further adjusted, according to the contribution of each channel to the classification accuracy via SVM, and the top channels with larger weights are eventually selected. The results show that the average accuracy of all subjects can reach 80.03% by using the first 30 channels with the largest weights from among the total of 118 channels. For the right hand and foot motor-imagery tasks, the generally applicable optimal channels are mostly located in the left hemisphere. Our generally applicable channel observation of the whole brain cortex suggests contralateral control correspondence: for unilateral motor imagery, the optimal channels are concentrated in the contralateral hemisphere. This is consistent with the contralateral control of the body by the human brain: the majority of the human motor and sensory fibers tend to control the contralateral limbs and pass through the midline of the body. Our proposed method provides optimal acquisition and analysis of the positions of EEG signals in specific motor-imagery tasks.

9 August 2021 in Journal of Healthcare Engineering
Verified
Journal Article
Open access
PDF file
QWQisong Wang
Qisong Wang
ZDZhening Dong
Zhening Dong
DLDan Liu
Dan Liu
Tian-ao CaoTian-ao Cao
Tian-ao Cao
MZMeiyan Zhang
Meiyan Zhang
JSJinwei Sun
Jinwei Sun
Respiratory diseases including apnea are often accompanied by abnormal respiratory depth, frequency, and rhythm. If different abnormal respiratory patterns can be detected and recorded, with their depth, frequency, and rhythm analyzed, the detection and diagnosis of respiratory diseases can be achieved. High-frequency millimeter-wave radar (76–81 GHz) has low environmental impact, high accuracy, and small volume, which is more suitable for respiratory signal detection and recognition compared with other contact equipment. In this paper, the experimental platform of frequency-modulated continuous wave (FMCW) radar was built at first, realizing the noncontact measurement of vital signs. Secondly, the energy intensity and threshold of respiration signal during each period were calculated by using the rectangular window, and the accurate judgment of apnea was realized via numerical comparison. Thirdly, the features of respiratory and heart rate signals, the number of peaks and valleys, the difference between peaks and valleys, the average and the standard deviation of normalized short-term energy, and the average and the standard deviation and the minimum of instantaneous frequency, were extracted and analyzed. Finally, support vector machine (SVM) and K-nearest neighbor (KNN) were used to classify the extracted features, and the accuracy was 98.25% and 88.75%, respectively. The classification and recognition of respiratory patterns have been successfully realized.

29 September 2025 in Frontiers in Neuroscience
Verified
Journal Article
Open access
PDF file
YDYiwei Dai
Yiwei Dai
ZCZhengkui Chen
Zhengkui Chen
Tian-ao CaoTian-ao Cao
Tian-ao Cao
HZHongyou Zhou
Hongyou Zhou
MFMin Fang
Min Fang
JTJijun Tong
Jijun Tong
Introduction: Steady-state visual evoked potential (SSVEP) has emerged as a pivotal branch in brain-computer interfaces (BCIs) due to its high signal-to-noise ratio (SNR) and elevated information transfer rate (ITR). However, substantial inter-subject variability in electroencephalographic (EEG) signals poses a significant challenge to current SSVEP frequency recognition. In particular, it is difficult to achieve high cross-subject classification accuracy in calibration-free scenarios, and the classification performance heavily depends on extensive calibration data.Methods: To mitigate the reliance on large calibration datasets and enhance cross-subject generalization, we propose SSVEP time-frequency fusion network (SSVEP-TFFNet), an improved deep learning network fusing time-domain and frequency-domain features dynamically. The network comprises two parallel branches: a time-domain branch that ingests raw EEG signals and a frequency-domain branch that processes complex-spectrum features. The two branches extract the time-domain and frequency-domain features, respectively. Subsequently, these features are fused via a dynamic weighting mechanism and input to the classifier. This fusion strategy strengthens the feature expression ability and generalization across different subjects.Results: Cross-subject classification was conducted on publicly available 12-class and 40-class SSVEP datasets. We also compared SSVEP-TFFNet with traditional approaches and principal deep learning methods. Results demonstrate that SSVEP-TFFNet achieves an average classification accuracy of 89.72% on the 12-class dataset, surpassing the best baseline method by 1.83%. SSVEP-TFFNet achieves average classification accuracies of 72.11 and 82.50% (40-class datasets), outperforming the best controlled method by 7.40 and 6.89% separately.Discussion: The performance validates the efficacy of dynamic time-frequency feature fusion and our proposed method provides a new paradigm for calibration-free SSVEP-based BCI systems.

22 August 2020 in Applied Sciences
Verified
Journal Article
Open access
PDF file
Tian-ao CaoTian-ao Cao
Tian-ao Cao
DLDan Liu
Dan Liu
QWQisong Wang
Qisong Wang
Ou BaiOu Bai
Ou Bai
JSJinwei Sun
Jinwei Sun
To improve the quality of lives of disabled people, the application of intelligent prosthesis was presented and investigated. In particular, surface Electromyography (sEMG) signals succeeded in controlling the manipulator in human–machine interface, due to the fact that EMG activity belongs to one of the most widely utilized biosignals and can reflect the straightforward motion intention of humans. However, the accuracy of real-time action recognition is usually low and there is usually obvious delay in a controlling manipulator, as a result of which the task of tracking human movement precisely, cannot be guaranteed. Therefore, this study proposes a method of action recognition and manipulator control. We built a multifunctional sEMG detection and action recognition system that integrated all discrete components. A biopotential measurement analog-to-digital converter with a high signal–noise rate (SNR) was chosen to ensure the high quality of the acquired sEMG signals. The acquired data were divided into sliding windows for processing in a shorter time. Mean Absolute Value (MAV), Waveform Length (WL), and Root Mean Square (RMS) were finally extracted and we found that compared to the Genetic-Algorithm-based Support Vector Machine (GA–SVM), the back propagation (BP) neural network performed better in joint action classification. The results showed that the average accuracy of judging the 5 actions (fist clenching, hand opening, wrist flexion, wrist extension, and calling me) was up to 93.2% and the response time was within 200 ms, which achieved a simultaneous control of the manipulator. Our work took into account the action recognition accuracy and real-time performance, and realized the sEMG-based manipulator control eventually, which made it easier for people with arm disabilities to communicate better with the outside world.

31 August 2022 in Sensors
Verified
Journal Article
Open access
PDF file
LTLin Tao
Lin Tao
Tian-ao CaoTian-ao Cao
Tian-ao Cao
QWQisong Wang
Qisong Wang
DLDan Liu
Dan Liu
JSJinwei Sun
Jinwei Sun
A brain-computer interface (BCI) translates a user’s thoughts such as motor imagery (MI) into the control of external devices. However, some people, who are defined as BCI illiteracy, cannot control BCI effectively. The main characteristics of BCI illiterate subjects are low classification rates and poor repeatability. To address the problem of MI-BCI illiteracy, we propose a distribution adaptation method based on multi-kernel learning to make the distribution of features between the source domain and target domain become even closer to each other, while the divisibility of categories is maximized. Inspired by the kernel trick, we adopted a multiple-kernel-based extreme learning machine to train the labeled source-domain data to find a new high-dimensional subspace that maximizes data divisibility, and then use multiple-kernel-based maximum mean discrepancy to conduct distribution adaptation to eliminate the difference in feature distribution between domains in the new subspace. In light of the high dimension of features of MI-BCI illiteracy, random forest, which can effectively handle high-dimensional features without additional cross-validation, was employed as a classifier. The proposed method was validated on an open dataset. The experimental results show that that the method we proposed suits MI-BCI illiteracy and can reduce the inter-domain differences, resulting in a reduction in the performance degradation of both cross-subjects and cross-sessions.

23 September 2025 in Symmetry
Verified
Journal Article
Open access
PDF file
Tian-ao CaoTian-ao Cao
Tian-ao Cao
HZHongyou Zhou
Hongyou Zhou
ZCZhengkui Chen
Zhengkui Chen
YDYiwei Dai
Yiwei Dai
MFMin Fang
Min Fang
JTJijun Tong
Jijun Tong
Electromyography (EMG) signals reflect hand motion intention and exhibit a certain degree of amplitude symmetry. Nowadays, recognition of hand motion intention based on EMG has enriched its burgeoning promotion in various applications, such as rehabilitation, prostheses, and intelligent supply chains. For instance, the motion intentions of humans can be conveyed to logistics equipment, thereby improving the level of intelligence in a supply chain. To enhance the recognition accuracy of multiple hand motion intentions, this paper proposes a hand motion intention recognition method that decodes EMG signals based on improved long short-term memory (LSTM). Firstly, we performed preprocessing and utilized overlapping sliding windows on EMG segments. Secondly, we chose LSTM and improved it so as to capture features and enable prediction of hand motion intention. Specifically, we introduced the optimal key hyperparameter combination in the LSTM model using a genetic algorithm (GA). We found that our proposed method achieved relatively high accuracy in detecting hand motion intention, with average accuracies of 92.0% (five gestures) and 89.7% (seven gestures), while the highest accuracy reached 100.0% (seven gestures). Our paper may provide a way to predict the motion intention of the human hand for intention communication.