Lei Guo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Lei Guo has a total of 6 co-authors. Here are the top co-authors by number of shared publications:
- Dapeng Niu
Northeastern University
3 shared publications - Jiaxu Zhao
College of Information Science and Engineering, Northeastern University, Shenyang, China
3 shared publications - Mingxing Jia
College of Information Science and Engineering, Northeastern University, Shenyang, China
3 shared publications Dapeng Niu
1 shared publicationMingxing Jia
1 shared publicationXingyu Han
College of Information Science and Engineering, Northeastern University, Shenyang, Liaoning, China
1 shared publication
Latest Publications
- 23 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 37 (6) , 2944-2958
MtvTrack: Robust Visual Tracking via Modeling Time-Variant State of the Target
Current single-object tracking algorithms depend on the information supplied by the template to identify and locate the object within the search area. However, environmental complexities and unknown factors can alter the object’s state, causing mismatches in template information. The existing works using the template update mechanism (TUM) and multiple template feature fusion have the following problems: 1) TUM is affected by input superposition, making it hard to eliminate noise; 2) they suffer a temporal lag in their responsiveness to changes that occur in the object during the tracking process; 3) it is insufficient to rely solely on visual features within the search area of the current frame to improve the template; and 4) the prior knowledge regarding the input is not fully leveraged to learn the time-variant state of the object. We observe that in complex tracking scenarios, humans subconsciously analyze the evolutionary patterns of the object and its surroundings and integrate this information with the object’s initial impression, thereby maintaining an awareness of the object’s temporal state. Motivated by this, we propose a novel solution to the above problem, named MtvTrack, which can model the time-variant state of the object through the dynamic evolution pattern and static initial impression. Simultaneously, we propose a method for predicting the evolution pattern of scenes by utilizing past, present, and future (PPF) states. This approach effectively eliminates the information redundancy between consecutive frames and addresses the issue of delayed predictions of the target state in relation to changes within the search area. We establish a joint probability generative model and fully utilize prior knowledge to learn the time-variant state of the object. In addition, we develop a vector quantized-PPF (VQPPF) module for predicting the object’s time-variant state. Experimental results on public benchmarks confirm the superior performance of our method. Source code is available at: https://github.com/long-wa/MtvTrack-main
- 23 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 37 (6) , 2944-2958
MtvTrack: Robust Visual Tracking via Modeling Time-Variant State of the Target
Current single-object tracking algorithms depend on the information supplied by the template to identify and locate the object within the search area. However, environmental complexities and unknown factors can alter the object’s state, causing mismatches in template information. The existing works using the template update mechanism (TUM) and multiple template feature fusion have the following problems: 1) TUM is affected by input superposition, making it hard to eliminate noise; 2) they suffer a temporal lag in their responsiveness to changes that occur in the object during the tracking process; 3) it is insufficient to rely solely on visual features within the search area of the current frame to improve the template; and 4) the prior knowledge regarding the input is not fully leveraged to learn the time-variant state of the object. We observe that in complex tracking scenarios, humans subconsciously analyze the evolutionary patterns of the object and its surroundings and integrate this information with the object’s initial impression, thereby maintaining an awareness of the object’s temporal state. Motivated by this, we propose a novel solution to the above problem, named MtvTrack, which can model the time-variant state of the object through the dynamic evolution pattern and static initial impression. Simultaneously, we propose a method for predicting the evolution pattern of scenes by utilizing past, present, and future (PPF) states. This approach effectively eliminates the information redundancy between consecutive frames and addresses the issue of delayed predictions of the target state in relation to changes within the search area. We establish a joint probability generative model and fully utilize prior knowledge to learn the time-variant state of the object. In addition, we develop a vector quantized-PPF (VQPPF) module for predicting the object’s time-variant state. Experimental results on public benchmarks confirm the superior performance of our method. Source code is available at: https://github.com/long-wa/MtvTrack-main
- 1 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ASME Transactions on Mechatronics
- Vol. 30 (6) , 5976-5985
Operation Condition Assessment for Elevators Based on Multidomain Generalization and Adaptation
The performance level and health condition of elevators seriously affect occupants' safety. Condition assessment plays an important role in the prognostics and health management of elevators. However, it has been found through our previous research that there is a serious domain shift problem in the process of elevator condition evaluation, which creates a severe influence on the evaluation model's generalization performance. Therefore, this article proposes a novel multidomain generalization and adaptation approach to address the domain shift problem in elevator condition assessment due to performance degradation caused by different factors. First, a multidomain generalization (DG) method introducing perturbation and gradient matching is developed to solve the dilemma of local optimality and overfitting problems in DG. Subsequently, the evaluation model that has acquired knowledge from the seen domains is split into backbone and classification head, and moved into the unseen domain for adaptation. Second, to deal with the high cost of adding data annotations and the lack of visibility of domain information, a special unsupervised multidomain adaptation (DA) method is designed to train the evaluation model on the unseen domains. Concretely, information entropy maximization, clustering pseudolabel cross entropy, and cross-domain feature alignment are employed as the total optimization objective for unsupervised adaptation of data on unseen domains. The validity and feasibility of the proposed method are verified by the elevator operation condition data obtained from the acquisition experiment. Finally, through ablation experiments and comparison experiments with state-of-the-art methods, our method has shown optimal results, while highlighting its necessity and advancement.
- 1 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ASME Transactions on Mechatronics
- Vol. 30 (6) , 5976-5985
Operation Condition Assessment for Elevators Based on Multidomain Generalization and Adaptation
The performance level and health condition of elevators seriously affect occupants' safety. Condition assessment plays an important role in the prognostics and health management of elevators. However, it has been found through our previous research that there is a serious domain shift problem in the process of elevator condition evaluation, which creates a severe influence on the evaluation model's generalization performance. Therefore, this article proposes a novel multidomain generalization and adaptation approach to address the domain shift problem in elevator condition assessment due to performance degradation caused by different factors. First, a multidomain generalization (DG) method introducing perturbation and gradient matching is developed to solve the dilemma of local optimality and overfitting problems in DG. Subsequently, the evaluation model that has acquired knowledge from the seen domains is split into backbone and classification head, and moved into the unseen domain for adaptation. Second, to deal with the high cost of adding data annotations and the lack of visibility of domain information, a special unsupervised multidomain adaptation (DA) method is designed to train the evaluation model on the unseen domains. Concretely, information entropy maximization, clustering pseudolabel cross entropy, and cross-domain feature alignment are employed as the total optimization objective for unsupervised adaptation of data on unseen domains. The validity and feasibility of the proposed method are verified by the elevator operation condition data obtained from the acquisition experiment. Finally, through ablation experiments and comparison experiments with state-of-the-art methods, our method has shown optimal results, while highlighting its necessity and advancement.
- 6 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-13
Operation Condition Assessment for Elevators Based on Deep Siamese Network and T-S Semi-Supervision Model
As a significant vertical means of transportation in daily life, the elevator’s operation condition and performance level are closely related to the passengers’ safety. However, there still lacks systematic and advanced method for the condition evaluation at present. Therefore, a semi-supervised two-stage deep learning network composed of feature selector and classifier is proposed to implement elevator condition assessment in this paper. Firstly, through the analysis of elevator mechanism and structure, vibration and noise indicators are selected as evaluation indexes to evaluate the elevator’s state during operation process. Secondly, a comparative experiment is designed to collect signal data that reflect different performance levels, and the time-frequency feature parameters that can comprehensively provide the operation process information are extracted from the indicators’ data. Additionally, with purpose of carrying out effective space conversion and explore the complex relationship between features and categories, a deep Siamese multi-head self-attention network is established to spatially map the extracted feature data, contributing to the clustering effect of data in high dimensional space. To deal with the difficult access to label data in the actual assessment process, the teacher-student (T-S) network is adopted to carry out the semi-supervised training through the mixture of labeled data obtained from designed experiment and unlabeled data. Moreover, the Soft-match is introduced to address the imbalance between quantity and quality of pseudo labels in semi-supervised training. Finally, different comparative experiments and ablation experiment are conducted to verify the effectiveness and feasibility of the proposed method on the test data set.
- 6 March 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-13
Operation Condition Assessment for Elevators Based on Deep Siamese Network and T-S Semi-Supervision Model
As a significant vertical means of transportation in daily life, the elevator’s operation condition and performance level are closely related to the passengers’ safety. However, there still lacks systematic and advanced method for the condition evaluation at present. Therefore, a semi-supervised two-stage deep learning network composed of feature selector and classifier is proposed to implement elevator condition assessment in this paper. Firstly, through the analysis of elevator mechanism and structure, vibration and noise indicators are selected as evaluation indexes to evaluate the elevator’s state during operation process. Secondly, a comparative experiment is designed to collect signal data that reflect different performance levels, and the time-frequency feature parameters that can comprehensively provide the operation process information are extracted from the indicators’ data. Additionally, with purpose of carrying out effective space conversion and explore the complex relationship between features and categories, a deep Siamese multi-head self-attention network is established to spatially map the extracted feature data, contributing to the clustering effect of data in high dimensional space. To deal with the difficult access to label data in the actual assessment process, the teacher-student (T-S) network is adopted to carry out the semi-supervised training through the mixture of labeled data obtained from designed experiment and unlabeled data. Moreover, the Soft-match is introduced to address the imbalance between quantity and quality of pseudo labels in semi-supervised training. Finally, different comparative experiments and ablation experiment are conducted to verify the effectiveness and feasibility of the proposed method on the test data set.