Hongmei Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Hongmei Wang has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
- Jingyu Wang
School of Astronautics, Northwestern Polytechnical University, Xi’an 710072, China
6 shared publications - Pengfei Huang
School of Astronautics, Northwestern Polytechnical University, Xian, China
3 shared publications - Qiaorong Wu
School of Astronautics, Northwestern Polytechnical University, Xi’an 710072, China
3 shared publications - Cheng Liang
School of Astronautics, Northwestern Polytechnical University, Xi’an 710072, China
2 shared publications Chenkai Li
School of Astronautics, Northwestern Polytechnical University, Xian, China
2 shared publications- Daming Zhou
School of Astronautics, Northwestern Polytechnical University, Xi’an 710072, China
2 shared publications
Latest Publications
- Open Access
Mutual Enhancement Fusion Tracker for Video-Level RGB-T Object Tracking
RGB and thermal visual tracking is a highly effective solution for complex environments. However, existing multimodal trackers still face significant limitations. They often apply uniform fusion weights across the entire image. They also ignore the distinct physical properties of different modalities and lack historical temporal modeling. To address these issues, we propose a novel video-level framework named Mutual Enhancement Fusion Tracker (MEFT). First, we introduce a modality-aware mutual enhancement module. This module explicitly transfers visual texture from the RGB modality to the thermal modality. Simultaneously, it transfers heat saliency from the thermal modality to the RGB modality. Second, we design an adaptive spatial gating network. This network generates normalized pixel-level weights to ensure an optimal combination of features at different locations. Finally, the fused representations are processed by a spatiotemporal sequence backbone. Extensive experiments on four benchmark datasets demonstrate that our method achieves competitive performance with state-of-the-art RGB-T trackers. For instance, our tracker achieves 60.3% SR on LasHeR and 67.5% MSR on RGBT234. MEFT significantly improves both tracking accuracy and overall reliability in a wide range of challenging real-world scenarios.
Open AccessMutual Enhancement Fusion Tracker for Video-Level RGB-T Object Tracking
RGB and thermal visual tracking is a highly effective solution for complex environments. However, existing multimodal trackers still face significant limitations. They often apply uniform fusion weights across the entire image. They also ignore the distinct physical properties of different modalities and lack historical temporal modeling. To address these issues, we propose a novel video-level framework named Mutual Enhancement Fusion Tracker (MEFT). First, we introduce a modality-aware mutual enhancement module. This module explicitly transfers visual texture from the RGB modality to the thermal modality. Simultaneously, it transfers heat saliency from the thermal modality to the RGB modality. Second, we design an adaptive spatial gating network. This network generates normalized pixel-level weights to ensure an optimal combination of features at different locations. Finally, the fused representations are processed by a spatiotemporal sequence backbone. Extensive experiments on four benchmark datasets demonstrate that our method achieves competitive performance with state-of-the-art RGB-T trackers. For instance, our tracker achieves 60.3% SR on LasHeR and 67.5% MSR on RGBT234. MEFT significantly improves both tracking accuracy and overall reliability in a wide range of challenging real-world scenarios.
- Open Access
Infrared and Visible Image Fusion Network Based on Self-Compensating Lightweight Convolution
Deep learning has significantly improved the quality of infrared and visible image fusion. However, existing mainstream deep fusion networks often come with complex architectures and a large number of parameters. While general lightweight techniques can effectively reduce model complexity, they often weaken feature interactions during the lightweighting process, resulting in the loss of complementary texture and thermal information in fused images and making it difficult to balance fusion performance and model efficiency. To address these issues, this paper constructs an infrared and visible image fusion network based on a self-compensating lightweight convolution mechanism, named LWC-DenseFuse. The core of the network lies in a self-compensating lightweight convolution module, which goes beyond conventional convolution replacement and explicitly addresses feature degradation introduced by lightweight design. The module decouples spatial and channel correlations of standard convolution through depthwise convolution and pointwise convolution, while incorporating a channel attention mechanism to adaptively enhance salient features. Additionally, channel shuffle technology is employed to promote information exchange between groups, thereby enhancing feature interaction and compensating for the loss of critical information caused by lightweight design. To further improve the representation capability of the lightweight network during optimization, a staged training strategy with progressive loss weighting is introduced. Experimental evaluations demonstrate that the proposed fusion network significantly reduces the number of model parameters while ensuring real-time inference performance. Meanwhile, it effectively alleviates the performance degradation typically associated with lightweight architectures, as evidenced by improvements in information entropy and visual fidelity.
Open AccessInfrared and Visible Image Fusion Network Based on Self-Compensating Lightweight Convolution
Deep learning has significantly improved the quality of infrared and visible image fusion. However, existing mainstream deep fusion networks often come with complex architectures and a large number of parameters. While general lightweight techniques can effectively reduce model complexity, they often weaken feature interactions during the lightweighting process, resulting in the loss of complementary texture and thermal information in fused images and making it difficult to balance fusion performance and model efficiency. To address these issues, this paper constructs an infrared and visible image fusion network based on a self-compensating lightweight convolution mechanism, named LWC-DenseFuse. The core of the network lies in a self-compensating lightweight convolution module, which goes beyond conventional convolution replacement and explicitly addresses feature degradation introduced by lightweight design. The module decouples spatial and channel correlations of standard convolution through depthwise convolution and pointwise convolution, while incorporating a channel attention mechanism to adaptively enhance salient features. Additionally, channel shuffle technology is employed to promote information exchange between groups, thereby enhancing feature interaction and compensating for the loss of critical information caused by lightweight design. To further improve the representation capability of the lightweight network during optimization, a staged training strategy with progressive loss weighting is introduced. Experimental evaluations demonstrate that the proposed fusion network significantly reduces the number of model parameters while ensuring real-time inference performance. Meanwhile, it effectively alleviates the performance degradation typically associated with lightweight architectures, as evidenced by improvements in information entropy and visual fidelity.
- 23 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Implicit Doubly Stochastic Graph-Based Subspace Clustering for Hyperspectral Band Selection
Band selection plays a critical role in reducing data redundancy of hyperspectral images, where subspace clustering-based methods have shown remarkable potential, owing to the effective extraction of low-dimensional representations. Despite conducting band grouping competently, they not only overlook preserving original affinity structures but also require a manual graph normalization step, which could lead to suboptimal performance. To solve these issues, we propose a novel band selection method termed Implicit Doubly Stochastic Graph based subspace Clustering (IDSGC), facilitating end-to-end band affinity graph construction. Firstly, a latent bipartite graph modeling (LBGM) module is designed to decompose the inherent band similarities into bipartite graph, while mitigating the interference of noise links. Secondly, a novel implicit learning mechanism (ILM) of doubly stochastic graph is presented to refine graph structure and low-dimensional representations simultaneously, which seamlessly integrates the LBGM with subspace learning. Finally, a probability-constrained affinity graph that is implicitly normalized during optimization is generated, which can be directly used for band clustering without additional post-processing. Extensive experiments demonstrate that IDSGC outperforms several state-of-the-art band selection methods.
- 23 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-16
Implicit Doubly Stochastic Graph-Based Subspace Clustering for Hyperspectral Band Selection
Band selection plays a critical role in reducing data redundancy of hyperspectral images, where subspace clustering-based methods have shown remarkable potential, owing to the effective extraction of low-dimensional representations. Despite conducting band grouping competently, they not only overlook preserving original affinity structures but also require a manual graph normalization step, which could lead to suboptimal performance. To solve these issues, we propose a novel band selection method termed Implicit Doubly Stochastic Graph based subspace Clustering (IDSGC), facilitating end-to-end band affinity graph construction. Firstly, a latent bipartite graph modeling (LBGM) module is designed to decompose the inherent band similarities into bipartite graph, while mitigating the interference of noise links. Secondly, a novel implicit learning mechanism (ILM) of doubly stochastic graph is presented to refine graph structure and low-dimensional representations simultaneously, which seamlessly integrates the LBGM with subspace learning. Finally, a probability-constrained affinity graph that is implicitly normalized during optimization is generated, which can be directly used for band clustering without additional post-processing. Extensive experiments demonstrate that IDSGC outperforms several state-of-the-art band selection methods.
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23 (1545598X) , 1-5
RFA-Net: A Feature Refining-Focused Network for Aircraft Fine-Grained Recognition
Fine-grained aircraft recognition is a crucial task in remote sensing scenarios, while the challenge is to distinguish the high visual similarity among different categories. However, the large variations within the same category of aircraft could cause a severe recognition accuracy decrease. To solve the problem, an aircraft fine-grained recognition approach based on the Refining-Focused Network (RFA-Net) is proposed. Firstly, the Multi-scale Feature Refinement Module is presented to extract the intra-class differential feature of aircraft, in which bidirectional channel connections is applied to fuse the feature in both global and local manner. Secondly, the Multi-branch Feature Focusing Module is established to obtain the fine-grained inter-class features, which could expand the receptive field to enhance the discriminative characteristic between similar aircraft. Finally, a novel loss function which called Aspect-Ratio-Approximated IoU is specifically designed to improve the aircraft fine-grained recognition accuracy, while the convergence of predicted bounding boxes could be expedited as well. Experimental results show that the proposed approach outperforms other competitors.
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 23 (1545598X) , 1-5
RFA-Net: A Feature Refining-Focused Network for Aircraft Fine-Grained Recognition
Fine-grained aircraft recognition is a crucial task in remote sensing scenarios, while the challenge is to distinguish the high visual similarity among different categories. However, the large variations within the same category of aircraft could cause a severe recognition accuracy decrease. To solve the problem, an aircraft fine-grained recognition approach based on the Refining-Focused Network (RFA-Net) is proposed. Firstly, the Multi-scale Feature Refinement Module is presented to extract the intra-class differential feature of aircraft, in which bidirectional channel connections is applied to fuse the feature in both global and local manner. Secondly, the Multi-branch Feature Focusing Module is established to obtain the fine-grained inter-class features, which could expand the receptive field to enhance the discriminative characteristic between similar aircraft. Finally, a novel loss function which called Aspect-Ratio-Approximated IoU is specifically designed to improve the aircraft fine-grained recognition accuracy, while the convergence of predicted bounding boxes could be expedited as well. Experimental results show that the proposed approach outperforms other competitors.
- Open Access
- 18 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 8248-8260
Remote Sensing Small Object Detection Based on Multicontextual Information Aggregation
Due to wide field of view and background confusion, remote sensing objects are small and densely packed, commonly used detection methods detecting small object are not satisfactory. In this article, we propose method multi-contextual information aggregation YOLO (MCIA-YOLO), combining three novel modules to effectively aggregate multi-contextual information across channels, depths and pixels. Firstly, the channel-spatial information aggregation (CSA) module assembles spatial global features pursuant to channel contextual information, increasing the density of key information. Secondly, the shallow-deep information sparse aggregation (SDSA) module applies sparse cross self-attention mechanism. By sparsely correlating long-range dependency information across different regions, the representation capability of small target is enhanced while removing redundant information. Thirdly, to enrich local multi-scale features and better identify dense targets, multi-scale weighted aggregation (MWA) module convolves multi-receptive field information and performs weighted fusion. Our method demonstrates satisfactory performance on dataset VisDrone2019, UAVDT and NWPU VHR-10, especially in small objects detection, surpassing several state-of-the-art methods
- 18 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 8248-8260
Open AccessRemote Sensing Small Object Detection Based on Multicontextual Information Aggregation
Due to wide field of view and background confusion, remote sensing objects are small and densely packed, commonly used detection methods detecting small object are not satisfactory. In this article, we propose method multi-contextual information aggregation YOLO (MCIA-YOLO), combining three novel modules to effectively aggregate multi-contextual information across channels, depths and pixels. Firstly, the channel-spatial information aggregation (CSA) module assembles spatial global features pursuant to channel contextual information, increasing the density of key information. Secondly, the shallow-deep information sparse aggregation (SDSA) module applies sparse cross self-attention mechanism. By sparsely correlating long-range dependency information across different regions, the representation capability of small target is enhanced while removing redundant information. Thirdly, to enrich local multi-scale features and better identify dense targets, multi-scale weighted aggregation (MWA) module convolves multi-receptive field information and performs weighted fusion. Our method demonstrates satisfactory performance on dataset VisDrone2019, UAVDT and NWPU VHR-10, especially in small objects detection, surpassing several state-of-the-art methods
- Open Access
- 1 January 2025
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 19 (1)
Infrared and Visible Image Fusion Based on Autoencoder Network
To overcome the problems of texture information loss and insufficiently prominent targets in existing fusion networks, an information decomposition‐based autoencoder fusion network for infrared and visible images is proposed in this paper. Two salient information encoders with unshared weights and two scene information encoders with shared weights are designed to extract different features from infrared and visible images, respectively. The constraint is added to the loss function in order to ensure the ability of the salient information encoders to extract representative features and the scene information encoder to extract the cross‐modality feature. In addition, by introducing the pre‐trained semantic segmentation networks to guide the network training and constructing a feature saliency‐based fusion strategy, the ability of the fusion network is further enhanced to distinguish between targets and backgrounds. Extensive experiments are carried out on five datasets. Comparison experiments with state‐of‐the‐art fusion networks and ablation experiments indicate that the proposed method can obtain fused images with richer and more comprehensive information and is more robust to challenging factors, such as strong and weak light smoke and fog environments. At the same time, the fused images by our proposed method are more beneficial for downstream tasks such as target detection.
- 1 January 2025
- Published by Institution of Engineering and Technology (IET) in IET Image Processing
- Vol. 19 (1)
Open AccessInfrared and Visible Image Fusion Based on Autoencoder Network
To overcome the problems of texture information loss and insufficiently prominent targets in existing fusion networks, an information decomposition‐based autoencoder fusion network for infrared and visible images is proposed in this paper. Two salient information encoders with unshared weights and two scene information encoders with shared weights are designed to extract different features from infrared and visible images, respectively. The constraint is added to the loss function in order to ensure the ability of the salient information encoders to extract representative features and the scene information encoder to extract the cross‐modality feature. In addition, by introducing the pre‐trained semantic segmentation networks to guide the network training and constructing a feature saliency‐based fusion strategy, the ability of the fusion network is further enhanced to distinguish between targets and backgrounds. Extensive experiments are carried out on five datasets. Comparison experiments with state‐of‐the‐art fusion networks and ablation experiments indicate that the proposed method can obtain fused images with richer and more comprehensive information and is more robust to challenging factors, such as strong and weak light smoke and fog environments. At the same time, the fused images by our proposed method are more beneficial for downstream tasks such as target detection.
- Open Access
- 26 November 2024
- Published by MDPI AG in Remote Sensing
- Vol. 16 (23) , 4420
An Improved DETR Based on Angle Denoising and Oriented Boxes Refinement for Remote Sensing Object Detection
Remote sensing image object detection presents significant challenges, due to the difficulty in accurately predicting the rotational angles of ground-oriented objects, coupled with issues such as the false and missed detection caused by insufficient object information. Moreover, traditional convolutional neural networks are inherently limited in their capacity to capture global contextual information. To address these challenges, a DETR-based remote sensing image object detection model is designed for oriented objects. Except for the backbone, transformer encoders and decoders, scenario query guiding modules, oriented boxes refinement modules, auxiliary multiple detectors, and oriented boxes denoising modules are also designed and included in our network. The scenario query guiding module is proposed that implicitly guides the decoder to focus more on object classification information specific to that scene when inferring. The multiple deformable attention mechanism is improved to the oriented one and utilized into the oriented boxes refinement module which repeatedly corrects the oriented boxes, enhancing the network’s ability to predict the oriented boxes precisely. At the same time, the improved auxiliary multiple detectors and oriented boxes denoising module are applied only for the training process to enhance the learning ability of the encoder and decoder for oriented objects. The ablation experiments proved the effectiveness of the designed modules. The detection accuracy of our network on DOTAv1.0 (76.77%) and HRCS2016 (97.01%) is improved compared with the state-of-the-art methods, which are especially significantly higher than DETR detection algorithms.
- 26 November 2024
- Published by MDPI AG in Remote Sensing
- Vol. 16 (23) , 4420
Open AccessAn Improved DETR Based on Angle Denoising and Oriented Boxes Refinement for Remote Sensing Object Detection
Remote sensing image object detection presents significant challenges, due to the difficulty in accurately predicting the rotational angles of ground-oriented objects, coupled with issues such as the false and missed detection caused by insufficient object information. Moreover, traditional convolutional neural networks are inherently limited in their capacity to capture global contextual information. To address these challenges, a DETR-based remote sensing image object detection model is designed for oriented objects. Except for the backbone, transformer encoders and decoders, scenario query guiding modules, oriented boxes refinement modules, auxiliary multiple detectors, and oriented boxes denoising modules are also designed and included in our network. The scenario query guiding module is proposed that implicitly guides the decoder to focus more on object classification information specific to that scene when inferring. The multiple deformable attention mechanism is improved to the oriented one and utilized into the oriented boxes refinement module which repeatedly corrects the oriented boxes, enhancing the network’s ability to predict the oriented boxes precisely. At the same time, the improved auxiliary multiple detectors and oriented boxes denoising module are applied only for the training process to enhance the learning ability of the encoder and decoder for oriented objects. The ablation experiments proved the effectiveness of the designed modules. The detection accuracy of our network on DOTAv1.0 (76.77%) and HRCS2016 (97.01%) is improved compared with the state-of-the-art methods, which are especially significantly higher than DETR detection algorithms.
- 21 August 2023
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 236
Online object tracking based interactive attention
- 21 August 2023
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 236
Online object tracking based interactive attention
- Open Access
- 24 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 78956-78969
Infrared and Visible Image Fusion Based on Autoencoder Composed of CNN-Transformer
Image fusion model based on autoencoder network gets more attention because it does not need to design fusion rules manually. However, most autoencoder-based fusion networks use two-stream CNNs with the same structure as the encoder, which are unable to extract global features due to the local receptive field of convolutional operations and lack the ability to extract unique features from infrared and visible images. A novel autoencoder-based image fusion network which consist of encoder module, fusion module and decoder module is constructed in this paper. For the encoder module, the CNN and Transformer are combined to capture the local and global feature of the source images simultaneously. In addition, novel contrast and gradient enhancement feature extraction blocks are designed respectively for infrared and visible images to maintain the information specific to each source images. The feature images obtained from encoder module are concatenated by the fusion module and input to the decoder module to obtain the fused image. Experimental results on three datasets show that the proposed network can better preserve both the clear target and detailed information of infrared and visible images respectively, and outperforms some state-of-the-art methods in both subjective and objective evaluation. At the same time, the fused image obtained by our proposed network can acquire the highest mean average precision in the target detection which proves that image fusion is beneficial for downstream tasks.
- 24 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 78956-78969
Open AccessInfrared and Visible Image Fusion Based on Autoencoder Composed of CNN-Transformer
Image fusion model based on autoencoder network gets more attention because it does not need to design fusion rules manually. However, most autoencoder-based fusion networks use two-stream CNNs with the same structure as the encoder, which are unable to extract global features due to the local receptive field of convolutional operations and lack the ability to extract unique features from infrared and visible images. A novel autoencoder-based image fusion network which consist of encoder module, fusion module and decoder module is constructed in this paper. For the encoder module, the CNN and Transformer are combined to capture the local and global feature of the source images simultaneously. In addition, novel contrast and gradient enhancement feature extraction blocks are designed respectively for infrared and visible images to maintain the information specific to each source images. The feature images obtained from encoder module are concatenated by the fusion module and input to the decoder module to obtain the fused image. Experimental results on three datasets show that the proposed network can better preserve both the clear target and detailed information of infrared and visible images respectively, and outperforms some state-of-the-art methods in both subjective and objective evaluation. At the same time, the fused image obtained by our proposed network can acquire the highest mean average precision in the target detection which proves that image fusion is beneficial for downstream tasks.