Yi Hou
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yi Hou has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Shilin Zhou
College of Electronic Science and Technology, National University of Defense Technology, Changsha, China
9 shared publications Shilin Zhou
College of Electronic Science and Technology National University of Defense Technology Changsha China
6 shared publications- Ye Zhang
Sun Yat-sen University
6 shared publications Hong Zhang
University of Alberta
5 shared publications- Kewei Ouyang
College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China
5 shared publications - Peibo Chen
College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China
3 shared publications
Latest Publications
- 23 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 37 (3) , 1277-1289
Trend and Order Features for Semi-Supervised Time-Series Classification via Multitask Learning
Multitask learning with a pretext task has excelled in time-series classification task lacking labeled data. The key to multitask learning is to build a pretext task and learn the most representative feature from the raw time series. In this article, we propose trend and order features for semi-supervised time-series classification via multitask learning (TOFL). Specifically, we propose a simple but effective pretext task—self-sequence order prediction (SOP)—to discover the order relation. In addition, we design a gradual trend fusion (GTF) block concatenating different trend features as the shared backbone network basis element to obtain high-quality trend features for the SOP task. Finally, we not only theoretically analyze the uniform stability and generalization error of TOFL but also evaluate the results compared with state-of-the-art (SOTA) supervised and semi-supervised methods on the 128 UCR datasets and three real-world datasets. TOFL demonstrates a high level of competitiveness and, in most cases, closely matches or even surpasses SOTA methods in terms of accuracy. The source code and data of TOFL are freely available at: https://github.com/Sample-design-alt/TOFL
- 23 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 37 (3) , 1277-1289
Trend and Order Features for Semi-Supervised Time-Series Classification via Multitask Learning
Multitask learning with a pretext task has excelled in time-series classification task lacking labeled data. The key to multitask learning is to build a pretext task and learn the most representative feature from the raw time series. In this article, we propose trend and order features for semi-supervised time-series classification via multitask learning (TOFL). Specifically, we propose a simple but effective pretext task—self-sequence order prediction (SOP)—to discover the order relation. In addition, we design a gradual trend fusion (GTF) block concatenating different trend features as the shared backbone network basis element to obtain high-quality trend features for the SOP task. Finally, we not only theoretically analyze the uniform stability and generalization error of TOFL but also evaluate the results compared with state-of-the-art (SOTA) supervised and semi-supervised methods on the 128 UCR datasets and three real-world datasets. TOFL demonstrates a high level of competitiveness and, in most cases, closely matches or even surpasses SOTA methods in terms of accuracy. The source code and data of TOFL are freely available at: https://github.com/Sample-design-alt/TOFL
- Open Access
Decoupled Early Time Series Classification Using Varied-Length Feature Augmentation and Gradient Projection Technique
Early time series classification (ETSC) is crucial for real-world time-sensitive applications. This task aims to classify time series data with least timestamps at the desired accuracy. Early methods used fixed-length time series to train the deep models, and then quit the classification process by setting specific exiting rules. However, these methods may not adapt to the length variation of flow data in ETSC. Recent advances have proposed end-to-end frameworks, which leveraged the Recurrent Neural Networks to handle the varied-length problems, and the exiting subnets for early quitting. Unfortunately, the conflict between the classification and early exiting objectives is not fully considered. To handle these problems, we decouple the ETSC task into the varied-length TSC task and the early exiting task. First, to enhance the adaptive capacity of classification subnets to the data length variation, a feature augmentation module based on random length truncation is proposed. Then, to handle the conflict between classification and early exiting, the gradients of these two tasks are projected into a unified direction. Experimental results on 12 public datasets demonstrate the promising performance of our proposed method.
Open AccessDecoupled Early Time Series Classification Using Varied-Length Feature Augmentation and Gradient Projection Technique
Early time series classification (ETSC) is crucial for real-world time-sensitive applications. This task aims to classify time series data with least timestamps at the desired accuracy. Early methods used fixed-length time series to train the deep models, and then quit the classification process by setting specific exiting rules. However, these methods may not adapt to the length variation of flow data in ETSC. Recent advances have proposed end-to-end frameworks, which leveraged the Recurrent Neural Networks to handle the varied-length problems, and the exiting subnets for early quitting. Unfortunately, the conflict between the classification and early exiting objectives is not fully considered. To handle these problems, we decouple the ETSC task into the varied-length TSC task and the early exiting task. First, to enhance the adaptive capacity of classification subnets to the data length variation, a feature augmentation module based on random length truncation is proposed. Then, to handle the conflict between classification and early exiting, the gradients of these two tasks are projected into a unified direction. Experimental results on 12 public datasets demonstrate the promising performance of our proposed method.
- RetractedOpen Access
- 29 September 2022
- Published by Wiley in Journal of Sensors
- Vol. 2022, 1-8
Determination of Evidence Weights Based on Convolutional Neural Network for Classification Problem
The weighted average is an efficient way to address conflicting evidence combination in the Dempster-Shafer evidence theory. However, it is an open issue how to determine the evidence weights reasonably. Although many traditional conflicting evidence combination solutions based on evidence distance or entropy have been put forward, the evidence weights are determined with a single aspect, and no comprehensive consideration of other useful information affects the weights. Thus, it does not ensure that determination of weights is the most reasonable. By introducing deep learning into conflicting evidence combination, this paper proposes a comprehensive method for determining the evidence weights based on a convolutional neural network. Taking the evidence as the network input and the corresponding weight as the output, it utilizes convolutional neural network to fully mine potentially useful information that affects the evidence weights, in order to determine the weights comprehensively. Additionally, we define a weight loss function. The weights are continuously optimized through back propagation and achieve the optimal when the weight loss function value is the minimum. Classification experimental results demonstrate that the proposed method outperforms traditional ones based on evidence distance or entropy and can be flexibly extended to other application fields as a decision-making fusion method.
- 29 September 2022
- Published by Wiley in Journal of Sensors
- Vol. 2022, 1-8
RetractedOpen AccessDetermination of Evidence Weights Based on Convolutional Neural Network for Classification Problem
The weighted average is an efficient way to address conflicting evidence combination in the Dempster-Shafer evidence theory. However, it is an open issue how to determine the evidence weights reasonably. Although many traditional conflicting evidence combination solutions based on evidence distance or entropy have been put forward, the evidence weights are determined with a single aspect, and no comprehensive consideration of other useful information affects the weights. Thus, it does not ensure that determination of weights is the most reasonable. By introducing deep learning into conflicting evidence combination, this paper proposes a comprehensive method for determining the evidence weights based on a convolutional neural network. Taking the evidence as the network input and the corresponding weight as the output, it utilizes convolutional neural network to fully mine potentially useful information that affects the evidence weights, in order to determine the weights comprehensively. Additionally, we define a weight loss function. The weights are continuously optimized through back propagation and achieve the optimal when the weight loss function value is the minimum. Classification experimental results demonstrate that the proposed method outperforms traditional ones based on evidence distance or entropy and can be flexibly extended to other application fields as a decision-making fusion method.
- Open Access
- 15 August 2022
- Published by MDPI AG in Information
- Vol. 13 (8) , 387
Logarithmic Negation of Basic Probability Assignment and Its Application in Target Recognition
The negation of probability distribution is a new perspective from which to obtain information. Dempster–Shafer (D–S) evidence theory, as an extension of possibility theory, is widely used in decision-making-level fusion. However, how to reasonably construct the negation of basic probability assignment (BPA) in D–S evidence theory is an open issue. This paper proposes a new negation of BPA, logarithmic negation. It solves the shortcoming of Yin’s negation that maximal entropy cannot be obtained when there are only two focal elements in the BPA. At the same time, the logarithmic negation of BPA inherits the good properties of the negation of probability, such as order reversal, involution, convergence, degeneration, and maximal entropy. Logarithmic negation degenerates into Gao’s negation when the values of the elements all approach 0. In addition, the data fusion method based on logarithmic negation has a higher belief value of the correct target in target recognition application.
- 15 August 2022
- Published by MDPI AG in Information
- Vol. 13 (8) , 387
Open AccessLogarithmic Negation of Basic Probability Assignment and Its Application in Target Recognition
The negation of probability distribution is a new perspective from which to obtain information. Dempster–Shafer (D–S) evidence theory, as an extension of possibility theory, is widely used in decision-making-level fusion. However, how to reasonably construct the negation of basic probability assignment (BPA) in D–S evidence theory is an open issue. This paper proposes a new negation of BPA, logarithmic negation. It solves the shortcoming of Yin’s negation that maximal entropy cannot be obtained when there are only two focal elements in the BPA. At the same time, the logarithmic negation of BPA inherits the good properties of the negation of probability, such as order reversal, involution, convergence, degeneration, and maximal entropy. Logarithmic negation degenerates into Gao’s negation when the values of the elements all approach 0. In addition, the data fusion method based on logarithmic negation has a higher belief value of the correct target in target recognition application.
- 29 April 2022
- Published by Springer Nature in Applied Intelligence
- Vol. 53 (2) , 1505-1516
Knowledge transfer via distillation from time and frequency domain for time series classification
- 29 April 2022
- Published by Springer Nature in Applied Intelligence
- Vol. 53 (2) , 1505-1516
Knowledge transfer via distillation from time and frequency domain for time series classification
- Open Access
- 28 November 2021
- Published by Institution of Engineering and Technology (IET) in IET Computer Vision
- Vol. 16 (3) , 255-265
Exploring complementary information of self‐supervised pretext tasks for unsupervised video pre‐training
This study addresses the problem of the unsupervised pre‐training of video representation learning. The authors' focus is on two common approaches: knowledge distillation and self‐supervised learning. The insight provided is that knowledge distillation and the rapidly advancing self‐supervised technique can be mutually beneficial. Combining these two approaches, a unified framework of self‐supervised learning and image‐based distillation (SSID) for unsupervised video pre‐training is proposed. The effectiveness of SSID in comparison to both image‐based distillation methods and the existing self‐supervised pre‐training baseline is demonstrated. In particular, the authors' model leverages three signals from the unlabelled data. First, the authors distil from the classifier of a 2D pre‐trained model as a soft label. To regularize the training process, the authors then build a novel positive pair of contrastive learning on the representation of the 2D/3D model. Finally, a self‐supervised pretext task is introduced to enhance the authors' model to become aware of the temporal evolution. The authors' experiment results showed that the learnt features achieved the best performance when transferred to action recognition tasks on UCF101 and HMDB51, reaching increases of 2.4% and 1.9% compared to the existing unsupervised pre‐training model, respectively.
- 28 November 2021
- Published by Institution of Engineering and Technology (IET) in IET Computer Vision
- Vol. 16 (3) , 255-265
Open AccessExploring complementary information of self‐supervised pretext tasks for unsupervised video pre‐training
This study addresses the problem of the unsupervised pre‐training of video representation learning. The authors' focus is on two common approaches: knowledge distillation and self‐supervised learning. The insight provided is that knowledge distillation and the rapidly advancing self‐supervised technique can be mutually beneficial. Combining these two approaches, a unified framework of self‐supervised learning and image‐based distillation (SSID) for unsupervised video pre‐training is proposed. The effectiveness of SSID in comparison to both image‐based distillation methods and the existing self‐supervised pre‐training baseline is demonstrated. In particular, the authors' model leverages three signals from the unlabelled data. First, the authors distil from the classifier of a 2D pre‐trained model as a soft label. To regularize the training process, the authors then build a novel positive pair of contrastive learning on the representation of the 2D/3D model. Finally, a self‐supervised pretext task is introduced to enhance the authors' model to become aware of the temporal evolution. The authors' experiment results showed that the learnt features achieved the best performance when transferred to action recognition tasks on UCF101 and HMDB51, reaching increases of 2.4% and 1.9% compared to the existing unsupervised pre‐training model, respectively.
- 1 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 312-317
A Lightweight Autonomous Vehicle System Based On Pure Visual Navigation
In recent years, autonomous driving solutions that rely on the fusion of multiple sensors such as lidar, inertial navigation systems, and GPS positioning systems have attracted the attention of the industry and academia. Such solutions use multiple sets of sensing systems to work together to ensure the greatest probability of the accuracy and completeness of the results. These solutions are able to tackle problems that are difficult to solve with a single sensor in the short term. However, in the long run, the design of deep coupling between data and strategy is not conducive to the real improvement of the perception system. Therefore, this paper proposes a light-weight autonomous vehicle system based on pure vision named VOLWA. To cope with the huge challenge of using the camera as the only sensor, VOLWA contains three new algorithms:(i) view segmentation algorithm is proposed to improve the perception accuracy (ii) visual place recognition algorithm for removing dynamic targets based on channel selection is proposed to solve the interference problem of dynamic targets such as pedestrians and vehicles in the field of view (iii) threshold self-localization based on image features is proposed for high-precision positioning of autonomous vehicles. Experimental results show that visual place recognition algorithm for removing dynamic targets based on channel selection effectively improves the performance of place recognition. The system architecture and algorithm proposed in this paper have been transplanted to self-developed autonomous vehicles for verification, and the autonomous vehicles are running well. See the link for the demo video https://youtu.be/6WfUnor205M.
- 1 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 312-317
A Lightweight Autonomous Vehicle System Based On Pure Visual Navigation
In recent years, autonomous driving solutions that rely on the fusion of multiple sensors such as lidar, inertial navigation systems, and GPS positioning systems have attracted the attention of the industry and academia. Such solutions use multiple sets of sensing systems to work together to ensure the greatest probability of the accuracy and completeness of the results. These solutions are able to tackle problems that are difficult to solve with a single sensor in the short term. However, in the long run, the design of deep coupling between data and strategy is not conducive to the real improvement of the perception system. Therefore, this paper proposes a light-weight autonomous vehicle system based on pure vision named VOLWA. To cope with the huge challenge of using the camera as the only sensor, VOLWA contains three new algorithms:(i) view segmentation algorithm is proposed to improve the perception accuracy (ii) visual place recognition algorithm for removing dynamic targets based on channel selection is proposed to solve the interference problem of dynamic targets such as pedestrians and vehicles in the field of view (iii) threshold self-localization based on image features is proposed for high-precision positioning of autonomous vehicles. Experimental results show that visual place recognition algorithm for removing dynamic targets based on channel selection effectively improves the performance of place recognition. The system architecture and algorithm proposed in this paper have been transplanted to self-developed autonomous vehicles for verification, and the autonomous vehicles are running well. See the link for the demo video https://youtu.be/6WfUnor205M.
- Open Access
- 25 June 2021
- Published by MDPI AG in Algorithms
- Vol. 14 (7) , 192
Convolutional Neural Network with an Elastic Matching Mechanism for Time Series Classification
Recently, some researchers adopted the convolutional neural network (CNN) for time series classification (TSC) and have achieved better performance than most hand-crafted methods in the University of California, Riverside (UCR) archive. The secret to the success of the CNN is weight sharing, which is robust to the global translation of the time series. However, global translation invariance is not the only case considered for TSC. Temporal distortion is another common phenomenon besides global translation in time series. The scale and phase changes due to temporal distortion bring significant challenges to TSC, which is out of the scope of conventional CNNs. In this paper, a CNN architecture with an elastic matching mechanism, which is named Elastic Matching CNN (short for EM-CNN), is proposed to address this challenge. Compared with the conventional CNN, EM-CNN allows local time shifting between the time series and convolutional kernels, and a matching matrix is exploited to learn the nonlinear alignment between time series and convolutional kernels of the CNN. Several EM-CNN models are proposed in this paper based on diverse CNN models. The results for 85 UCR datasets demonstrate that the elastic matching mechanism effectively improves CNN performance.
- 25 June 2021
- Published by MDPI AG in Algorithms
- Vol. 14 (7) , 192
Open AccessConvolutional Neural Network with an Elastic Matching Mechanism for Time Series Classification
Recently, some researchers adopted the convolutional neural network (CNN) for time series classification (TSC) and have achieved better performance than most hand-crafted methods in the University of California, Riverside (UCR) archive. The secret to the success of the CNN is weight sharing, which is robust to the global translation of the time series. However, global translation invariance is not the only case considered for TSC. Temporal distortion is another common phenomenon besides global translation in time series. The scale and phase changes due to temporal distortion bring significant challenges to TSC, which is out of the scope of conventional CNNs. In this paper, a CNN architecture with an elastic matching mechanism, which is named Elastic Matching CNN (short for EM-CNN), is proposed to address this challenge. Compared with the conventional CNN, EM-CNN allows local time shifting between the time series and convolutional kernels, and a matching matrix is exploited to learn the nonlinear alignment between time series and convolutional kernels of the CNN. Several EM-CNN models are proposed in this paper based on diverse CNN models. The results for 85 UCR datasets demonstrate that the elastic matching mechanism effectively improves CNN performance.
- Open Access
- 17 June 2021
- Published by MDPI AG in Information
- Vol. 12 (6) , 252
Adaptive Multi-Scale Wavelet Neural Network for Time Series Classification
Wavelet transform is a well-known multi-resolution tool to analyze the time series in the time-frequency domain. Wavelet basis is diverse but predefined by manual without taking the data into the consideration. Hence, it is a great challenge to select an appropriate wavelet basis to separate the low and high frequency components for the task on the hand. Inspired by the lifting scheme in the second-generation wavelet, the updater and predictor are learned directly from the time series to separate the low and high frequency components of the time series. An adaptive multi-scale wavelet neural network (AMSW-NN) is proposed for time series classification in this paper. First, candidate frequency decompositions are obtained by a multi-scale convolutional neural network in conjunction with a depthwise convolutional neural network. Then, a selector is used to choose the optimal frequency decomposition from the candidates. At last, the optimal frequency decomposition is fed to a classification network to predict the label. A comprehensive experiment is performed on the UCR archive. The results demonstrate that, compared with the classical wavelet transform, AMSW-NN could improve the performance based on different classification networks.
- 17 June 2021
- Published by MDPI AG in Information
- Vol. 12 (6) , 252
Open AccessAdaptive Multi-Scale Wavelet Neural Network for Time Series Classification
Wavelet transform is a well-known multi-resolution tool to analyze the time series in the time-frequency domain. Wavelet basis is diverse but predefined by manual without taking the data into the consideration. Hence, it is a great challenge to select an appropriate wavelet basis to separate the low and high frequency components for the task on the hand. Inspired by the lifting scheme in the second-generation wavelet, the updater and predictor are learned directly from the time series to separate the low and high frequency components of the time series. An adaptive multi-scale wavelet neural network (AMSW-NN) is proposed for time series classification in this paper. First, candidate frequency decompositions are obtained by a multi-scale convolutional neural network in conjunction with a depthwise convolutional neural network. Then, a selector is used to choose the optimal frequency decomposition from the candidates. At last, the optimal frequency decomposition is fed to a classification network to predict the label. A comprehensive experiment is performed on the UCR archive. The results demonstrate that, compared with the classical wavelet transform, AMSW-NN could improve the performance based on different classification networks.
- Open Access
Conflict Management for Target Recognition Based on PPT Entropy and Entropy Distance
Conflicting evidence affects the final target recognition results. Thus, managing conflicting evidence efficiently can help to improve the belief degree of the true target. In current research, the existing approaches based on belief entropy use belief entropy itself to measure evidence conflict. However, it is not convincing to characterize the evidence conflict only through belief entropy itself. To solve this problem, we comprehensively consider the influences of the belief entropy itself and mutual belief entropy on conflict measurement, and propose a novel approach based on an improved belief entropy and entropy distance. The improved belief entropy based on pignistic probability transformation function is named pignistic probability transformation (PPT) entropy that measures the conflict between evidences from the perspective of self-belief entropy. Compared with the state-of-the-art belief entropy, it can measure the uncertainty of evidence more accurately, and make full use of the intersection information of evidence to estimate the degree of evidence conflict more reasonably. Entropy distance is a new distance measurement method and is used to measure the conflict between evidences from the perspective of mutual belief entropy. Two measures are mutually complementary in a sense. The results of numerical examples and target recognition applications demonstrate that our proposed approach has a faster convergence speed, and a higher belief degree of the true target compared with the existing methods.
Open AccessConflict Management for Target Recognition Based on PPT Entropy and Entropy Distance
Conflicting evidence affects the final target recognition results. Thus, managing conflicting evidence efficiently can help to improve the belief degree of the true target. In current research, the existing approaches based on belief entropy use belief entropy itself to measure evidence conflict. However, it is not convincing to characterize the evidence conflict only through belief entropy itself. To solve this problem, we comprehensively consider the influences of the belief entropy itself and mutual belief entropy on conflict measurement, and propose a novel approach based on an improved belief entropy and entropy distance. The improved belief entropy based on pignistic probability transformation function is named pignistic probability transformation (PPT) entropy that measures the conflict between evidences from the perspective of self-belief entropy. Compared with the state-of-the-art belief entropy, it can measure the uncertainty of evidence more accurately, and make full use of the intersection information of evidence to estimate the degree of evidence conflict more reasonably. Entropy distance is a new distance measurement method and is used to measure the conflict between evidences from the perspective of mutual belief entropy. Two measures are mutually complementary in a sense. The results of numerical examples and target recognition applications demonstrate that our proposed approach has a faster convergence speed, and a higher belief degree of the true target compared with the existing methods.