XDXue-Mei Dong

Xue-Mei Dong

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Xue-Mei Dong has a total of 33 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Attentive Eraser: Unleashing Diffusion Model’s Object Removal Potential via Self-Attention Redirection Guidance

    by Wenhao Sun, Xue-Mei Dong, Benlei Cui, Jingqun Tang

    Recently, diffusion models have emerged as promising newcomers in the field of generative models, shining brightly in image generation. However, when employed for object removal tasks, they still encounter issues such as generating random artifacts and the incapacity to repaint foreground object areas with appropriate content after removal. To tackle these problems, we propose Attentive Eraser, a tuning-free method to empower pre-trained diffusion models for stable and effective object removal. Firstly, in light of the observation that the self-attention maps influence the structure and shape details of the generated images, we propose Attention Activation and Suppression (ASS), which re-engineers the self-attention mechanism within the pre-trained diffusion models based on the given mask, thereby prioritizing the background over the foreground object during the reverse generation process. Moreover, we introduce Self-Attention Redirection Guidance (SARG), which utilizes the self-attention redirected by ASS to guide the generation process, effectively removing foreground objects within the mask while simultaneously generating content that is both plausible and coherent. Experiments demonstrate the stability and effectiveness of Attentive Eraser in object removal across a variety of pre-trained diffusion models, outperforming even training-based methods. Furthermore, Attentive Eraser can be implemented in various diffusion model architectures and checkpoints, enabling excellent scalability.


  • Open Access

    SCASNet: Spatial Context-Aware Selection Network for Small Object Detection in Aerial Imagery

    by Zhenkuan Wang, Xue-Mei Dong, Yongli Xu

    The detection of small objects within intricate backgrounds poses a significant challenge in the domain of aerial image object detection. In this manuscript, a spatial context-aware selection network (SCASNet) is proposed, which innovatively integrates a state space model with the YOLO architecture to address this challenge. A spatial selection block and a context-aware block are designed to form a spatial context-aware selection module, which can overcome the limitations of the original state space model in sequence modeling, such as insufficient receptive fields and weak local dependency modeling. Then, a channel prior multidimensional attention enhancement module is proposed to focus on key information and optimize the extraction of spatial relationships. It leverages multi-scale strip convolutions to map spatial relationships and dynamically allocates weights across channel and spatial dimensions. Finally, a content-focused attention module is designed in the detection heads to fuse fine-grained features from the lower layers of the backbone network with semantic features from the neck layers, which enhances the richness of feature representation. Extensive experiments conducted on publicly available datasets, VisDrone, AI-TOD and SSDD, demonstrate the competitive performance of the proposed SCASNet compared with existing aerial image object detection models.


  • Top Cited Papers

    DBCTNet: Double Branch Convolution-Transformer Network for Hyperspectral Image Classification

    by Rui Xu, Xue-Mei Dong, Weijie Li, Jiangtao Peng, Weiwei Sun, Yi Xu

    Currently, deep learning methods represented by convolutional neural networks (CNNs) or Transformers are of great interest in hyperspectral image (HSI) classification. And recent works show that hybrid models using CNN and Transformer modules are expected to achieve better performance than when they are used alone. However, these hybrid models applied to HSI classification consider the combination of 2D CNN and Transformer, which makes the models have high computational complexity. And the information of multiple spectral dimensions different from ordinary RGB images has not been fully excavated. Based on this, we propose DBCTNet, a double branch Convolution-Transformer network. Specifically, a MSpeFE module is used for multiscale spectral feature extraction at the early stage of the proposed network. Then a ConvTE block is designed to improve the original Transformer encoder, where a Conv spectral projection unit and a convolutional multihead self-attention (CMHSA) unit are proposed to extract spatial and global spectral features. A double branch module is further built based on 3D CNN and ConvTE. This module can fully integrate spatial and local-global spectral features, while also having low computational complexity. Experiment results on four public datasets, Pavia University, Houston, WHU-Hi-LongKou and HuangHeKou, show that DBCTNet achieves satisfactory performance with a small number of parameters and relatively excellent efficiency compared to nine other networks. The implement of DBCTNet will be available publicly at https://github.com/xurui-joei/DBCTNet.


  • A Knowledge Graph Reasoning Approach Integrating Attention-based LSTM and Multi-Agent Reinforcement Learning

    by Qingqing Wang, Han Jiao, Danpu Zhang, Xuemei Dong

    Knowledge reasoning methods play a pivotal role in various applications, including knowledge graph completion, knowledge-based question answering, and knowledge recommendation. Among these methods, path-based multi-hop reasoning techniques have the ability to leverage the rich graph information in knowledge graphs beyond triplets, but they still encounter certain challenges. Existing multi-hop knowledge reasoning methods heavily rely on data and lack interpretability. Additionally, the vast exploration space of paths, composed of numerous entities and relations in large knowledge graphs, often leads to irrelevant and redundant exploration. To address these challenges, this paper proposes a novel knowledge reasoning method named ALMARL (Attention-based LSTM and Multi-Agent Reinforcement Learning for Knowledge Graph Reasoning). It utilizes Attention-based LSTM in conjunction with multi-agent reinforcement learning. The method first employs clustering techniques to group entities. Based on the clustering results, multi-agents at different levels are established to selectively explore certain clusters or limit the search to specific clusters, effectively reducing the exploration of irrelevant entities and minimizing redundant exploration. Subsequently, the agents efficiently explore paths and effectively mine deep semantic information in entity relationships through the integration of Attention-based LSTM. Finally, the model produces inference results and extracts interpretable paths. We evaluate our proposed model on two types of tasks: link prediction and fact prediction. Experimental results demonstrate significant performance improvements compared to several competitive baselines.

  • ESSINet: Efficient Spatial–Spectral Interaction Network for Hyperspectral Image Classification

    by Zhuwang Lv, Xue-Mei Dong, Jiangtao Peng, Weiwei Sun

    Nowadays, convolutional neural networks (CNNs) are widely used in the field of hyperspectral image (HSI) classification. However, a major feature of HSIs is their rich spectral–spatial information with hundreds of continuous bands. This inevitably incurs the problems of high computational cost for network optimization and high interredundancy in the convolution kernels. To solve these problems, in this article, we rethink HSIs from the spectral perspective and introduce a lightweight operator called involution, which can effectively solve the above limitations. Different from traditional convolution kernels, the involution kernels pay more attention to the features of the channels but usually ignore the spatial features in the receptive field. To incorporate both spatial and spectral information, we construct a dual-pooling layer and design a novel involution-2D operator and its more lightweight version, involution-1D operator. Finally, an efficient spatial–spectral interaction network (ESSINet) for HSI classification is proposed based on these two new operators, which can make the spatial–spectral information in HSIs interact more closely. Extensive experimental results on four public datasets demonstrate the effectiveness and efficiency of the proposed ESSINet over some state-of-the-art CNN-based networks.


  • Open Access

    LiteSCANet: An Efficient Lightweight Network Based on Spectral and Channel-Wise Attention for Hyperspectral Image Classification

    by Su Qiao, Xue-Mei Dong, Jiangtao Peng, Weiwei Sun

    Deep learning (DL) algorithms have been demonstrated to have great potential in hyperspectral image (HSI) classification. However, most DL methods mainly focus on improving classification performance, neglecting the computational cost. In order to broaden the application scenarios of DL-based HSI classification methods, it is necessary to develop lightweight and fast models to fit the deployment needs of computational-resource-limited platforms. Taking into account the efficiency and accuracy, this article designs a lightweight network architecture based on spectral and channel-wise attention modules, namely LiteSCANet, for HSI classification. It contains a residual double-branch structure, which makes the model effectively extract spectral-spatial features and achieve good performance with fast inference speed and low computational consumption (i.e., floating-point operations, number of parameters, and graphics processing unit memory usage). The experiment results on four benchmark datasets show that our proposed model achieves an excellent trade-off between efficiency and accuracy compared with the other six existing networks.


  • Real-time Power Angle Measurement based on Second-Order Generalized Integrator with Frequency-Locked Loop

    by Xuemei Dong, Rende Zhao, Yansong Wang, Hailiang Xu

    The power angle of the synchronous generator is a key indicator that reflects the transient stability of the power system. The accurate measurement of the power angle is thus important to ensure the stable operation of the power system. In this paper, a real-time power angle measurement of synchronous generator is proposed, which is based on the second-order generalized integrator with frequency-locked loop (SOGI-FLL). Firstly, the terminal voltage and current of synchronous generator (SG) are collected in real-time. Secondly, the fundamental-frequency positive sequence signal is extracted from the instantaneous sampling data by using the SOGI-FLL. Then the generator vector equation is used to calculate the power angle. Meanwhile, the power angle error existing in the transient process is compensated by using the generator swing equation. The one machine infinite bus (OMIB) system is considered when building the power angle monitoring model of SG. Two scenarios under both steady state and transient state are respectively simulated with the MATLAB/Simulink. The simulation results indicate that the proposed method can enable a better real-time calculation with high accuracy for both scenarios.

  • LiteDepthwiseNet: A Lightweight Network for Hyperspectral Image Classification

    by Benlei Cui, Xue-Mei Dong, Qiaoqiao Zhan, Jiangtao Peng, Weiwei Sun

    Deep learning methods have shown considerable potential for hyperspectral image (HSI) classification, which can achieve high accuracy compared with traditional methods. However, they often need a large number of training samples and have a lot of parameters and high computational overhead. To solve these problems, this article proposes new network architecture, LiteDepthwiseNet, for HSI classification. Based on 3-D depthwise convolution, LiteDepthwiseNet can decompose standard convolution into depthwise convolution and pointwise convolution, which can achieve high classification performance with minimal parameters. Moreover, we remove the ReLU layer and batch normalization layer in the original 3-D depthwise convolution, which is likely to improve the overfitting phenomenon of the model on small-sized data sets. In addition, focal loss is used as the loss function to improve the model’s attention on difficult samples and unbalanced data, and its training performance is significantly better than that of cross-entropy loss or balanced cross-entropy loss. Experiment results on five benchmark hyperspectral data sets show that LiteDepthwiseNet achieves state-of-the-art performance with a very small number of parameters and low computational cost.


  • Open Access

    LiteDepthwiseNet: An Extreme Lightweight Network for Hyperspectral Image Classification

    by Benlei Cui, XueMei Dong, Qiaoqiao Zhan, Jiangtao Peng, Weiwei Sun

    Deep learning methods have shown considerable potential for hyperspectral image (HSI) classification, which can achieve high accuracy compared with traditional methods. However, they often need a large number of training samples and have a lot of parameters and high computational overhead. To solve these problems, this paper proposes a new network architecture, LiteDepthwiseNet, for HSI classification. Based on 3D depthwise convolution, LiteDepthwiseNet can decompose standard convolution into depthwise convolution and pointwise convolution, which can achieve high classification performance with minimal parameters. Moreover, we remove the ReLU layer and Batch Normalization layer in the original 3D depthwise convolution, which significantly improves the overfitting phenomenon of the model on small sized datasets. In addition, focal loss is used as the loss function to improve the model's attention on difficult samples and unbalanced data, and its training performance is significantly better than that of cross-entropy loss or balanced cross-entropy loss. Experiment results on three benchmark hyperspectral datasets show that LiteDepthwiseNet achieves state-of-the-art performance with a very small number of parameters and low computational cost.