Haitao Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Haitao Xu has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
Jie Gao
4 shared publicationsJinsong Du
4 shared publicationsWei Wang
4 shared publications- Jiangtao Xu
Tsinghua University
2 shared publications Xu Yang
2 shared publicationsJiang Sun
Aviation Industry Qing’an Group Co., Ltd, Xian, China
1 shared publication
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (4) , 9872-9884
GC-DETR: Geometry–Color Decoupled Transformer for 3-D Indoor Object Detection in Smart Consumer Devices
3D indoor object detection from point clouds is a core enabling technology for perceptive consumer systems such as service robots, smart home devices, and AR/VR terminals. Although geometric (XYZ) and color (RGB) information are both crucial for reliable on-device scene understanding, most existing approaches simply fuse them in a unified feature space, ignoring their heterogeneous characteristics and causing feature interference. We propose GC-DETR, a Geometry–Color Decoupled and Fused DEtection TRansformer designed to address modal interference in complex indoor environments. GC-DETR follows a decouple–then–fuse paradigm with three key components: a Geometry–Color Disentanglement Module (GCDM), a Dual-branch Feature Pyramid Network (DB-FPN), and a Geometry-guided Cross-Attention Module (G-CAM). Extensive experiments on ScanNetV2 show that GC-DETR achieves state-of-the-art performance, with 78.1% AP@0.25 and 65.1% AP@0.5. Cross-dataset evaluation on SUN RGB-D further demonstrates strong generalization and robustness. Additional ablation and edge-device deployment analyses further support the effectiveness and practical relevance of the proposed design. These results indicate that GC-DETR provides a robust and accurate solution for intelligent perception in next-generation consumer applications.
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (4) , 9872-9884
GC-DETR: Geometry–Color Decoupled Transformer for 3-D Indoor Object Detection in Smart Consumer Devices
3D indoor object detection from point clouds is a core enabling technology for perceptive consumer systems such as service robots, smart home devices, and AR/VR terminals. Although geometric (XYZ) and color (RGB) information are both crucial for reliable on-device scene understanding, most existing approaches simply fuse them in a unified feature space, ignoring their heterogeneous characteristics and causing feature interference. We propose GC-DETR, a Geometry–Color Decoupled and Fused DEtection TRansformer designed to address modal interference in complex indoor environments. GC-DETR follows a decouple–then–fuse paradigm with three key components: a Geometry–Color Disentanglement Module (GCDM), a Dual-branch Feature Pyramid Network (DB-FPN), and a Geometry-guided Cross-Attention Module (G-CAM). Extensive experiments on ScanNetV2 show that GC-DETR achieves state-of-the-art performance, with 78.1% AP@0.25 and 65.1% AP@0.5. Cross-dataset evaluation on SUN RGB-D further demonstrates strong generalization and robustness. Additional ablation and edge-device deployment analyses further support the effectiveness and practical relevance of the proposed design. These results indicate that GC-DETR provides a robust and accurate solution for intelligent perception in next-generation consumer applications.
- Open Access
- 24 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 780-784
Data-Driven Estimation of Respiratory Mechanics using Ventilator Waveforms and Thoracoabdominal Motion
- 24 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 780-784
Open AccessData-Driven Estimation of Respiratory Mechanics using Ventilator Waveforms and Thoracoabdominal Motion
- 9 October 2025
- Published by SAGE Publications in Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture
- Vol. 240 (10) , 1937-1950
Physics-informed graph convolutional neural networks with weighted adaptive loss for tool wear prediction under variable working conditions
Tool wear prediction is crucial for enhancing safety in industrial production environments and improving product quality. However, data-driven models often fail to effectively capture local wear features, physics-based models typically neglect real-time environmental fluctuations, and current fusion methods struggle to optimally balance the contributions of physics-based and data-driven components. To address these limitations, this paper introduces a novel physics-informed graph convolutional neural network (PGCN-WL) for tool wear prediction. This method incorporates a graph convolutional neural network (GCN), where extracted features serve as nodes in an undirected graph structure. GCN facilitates the continuous aggregation of feature information from neighboring nodes, iteratively refining node features to enhance the model’s sensitivity to local wear dynamics within the input signal. Furthermore, a multi-layer perceptron learns an adaptive loss function designed to complement the physics-informed loss term. This component specifically captures the dynamic evolution of cutting parameters during the wear process. An adaptive weight update mechanism is also proposed to dynamically adjust the influence of different loss functions, thereby improving overall prediction accuracy. Finally, the Model-Agnostic Meta-Learning (MAML) framework is applied. It treats data originating from distinct operational conditions as separate learning tasks, enabling the model to acquire versatile internal representations that can adapt swiftly to various working environment. The efficacy of this combined approach is validated using a publicly available tool wear prediction dataset, demonstrating its effectiveness.
- 9 October 2025
- Published by SAGE Publications in Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture
- Vol. 240 (10) , 1937-1950
Physics-informed graph convolutional neural networks with weighted adaptive loss for tool wear prediction under variable working conditions
Tool wear prediction is crucial for enhancing safety in industrial production environments and improving product quality. However, data-driven models often fail to effectively capture local wear features, physics-based models typically neglect real-time environmental fluctuations, and current fusion methods struggle to optimally balance the contributions of physics-based and data-driven components. To address these limitations, this paper introduces a novel physics-informed graph convolutional neural network (PGCN-WL) for tool wear prediction. This method incorporates a graph convolutional neural network (GCN), where extracted features serve as nodes in an undirected graph structure. GCN facilitates the continuous aggregation of feature information from neighboring nodes, iteratively refining node features to enhance the model’s sensitivity to local wear dynamics within the input signal. Furthermore, a multi-layer perceptron learns an adaptive loss function designed to complement the physics-informed loss term. This component specifically captures the dynamic evolution of cutting parameters during the wear process. An adaptive weight update mechanism is also proposed to dynamically adjust the influence of different loss functions, thereby improving overall prediction accuracy. Finally, the Model-Agnostic Meta-Learning (MAML) framework is applied. It treats data originating from distinct operational conditions as separate learning tasks, enabling the model to acquire versatile internal representations that can adapt swiftly to various working environment. The efficacy of this combined approach is validated using a publicly available tool wear prediction dataset, demonstrating its effectiveness.
- 17 March 2025
- Published by Elsevier in Journal of Power Sources
- Vol. 640
Feature selection of battery capacity estimation method based on deep learning
- 17 March 2025
- Published by Elsevier in Journal of Power Sources
- Vol. 640
Feature selection of battery capacity estimation method based on deep learning
- 14 December 2024
- Published by Elsevier in Aerospace Science and Technology
- Vol. 158
Data-driven simulation of aero-engine rotor assembly process optimization for unbalance based on fusion algorithm
- 14 December 2024
- Published by Elsevier in Aerospace Science and Technology
- Vol. 158
Data-driven simulation of aero-engine rotor assembly process optimization for unbalance based on fusion algorithm
- 7 September 2023
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 34 (12) , 125026
A novel pre-trained model based on graph-labeling graph neural networks for tool wear prediction under variable working conditions
In recent years, deep learning has made significant strides and found extensive applications in tool wear prediction. However, most methods based on deep learning rely on large-scale datasets for training and do not consider the method under variable working conditions. To address these limitations, this paper proposes a novel pre-trained fine-tuning model based on the graph-labeling graph neural network (GGPT) specifically designed to train on small-scale datasets. The GGPT model comprises two essential components: a feature extractor and a graph attention module (GAT). The feature extractor uses a residual connection constructed by maximum pooling and average pooling to improve the reliability of local feature learning. The GAT module enhances the global feature learning ability of the model by aggregating local features extracted by the feature extractor. In addition, Improved Pseudo-task Regularization is introduced to enable the model to converge better on small datasets. Experimental results substantiate the superiority of the GGPT model over other deep learning-based methods, particularly when confronted with variable working conditions for tool wear prediction.
- 7 September 2023
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 34 (12) , 125026
A novel pre-trained model based on graph-labeling graph neural networks for tool wear prediction under variable working conditions
In recent years, deep learning has made significant strides and found extensive applications in tool wear prediction. However, most methods based on deep learning rely on large-scale datasets for training and do not consider the method under variable working conditions. To address these limitations, this paper proposes a novel pre-trained fine-tuning model based on the graph-labeling graph neural network (GGPT) specifically designed to train on small-scale datasets. The GGPT model comprises two essential components: a feature extractor and a graph attention module (GAT). The feature extractor uses a residual connection constructed by maximum pooling and average pooling to improve the reliability of local feature learning. The GAT module enhances the global feature learning ability of the model by aggregating local features extracted by the feature extractor. In addition, Improved Pseudo-task Regularization is introduced to enable the model to converge better on small datasets. Experimental results substantiate the superiority of the GGPT model over other deep learning-based methods, particularly when confronted with variable working conditions for tool wear prediction.