JHJunqing Huang

Junqing Huang

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Junqing Huang has a total of 24 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Top Cited Papers
    Open Access

    GLMamba: A Global–Local Mamba Network for Efficient Remote Sensing Change Detection

    by Shengyan Liu, Chengye Zhu, Haoyu Yin, Kaibo Qin, Haifeng Lin, Junqing Huang, Min Xia, Liguo Weng

    With the rapid advancement of deep learning, Convolutional Neural Networks (CNNs) and Transformers have become two mainstream architectures for remote sensing change detection (CD). CNNs are effective at extracting local features but struggle to model long-range dependencies due to limited receptive fields. Transformers capture global contextual information through self-attention, yet their high computational cost limits their practicality for high-resolution remote sensing imagery. To balance global modeling capability and local detail representation, this paper introduces the Mamba architecture into CD and proposes Global–Locxnkal Mamba (GLMamba), an efficient network capable of jointly modeling long-range dependencies and fine-grained spatial details. Built upon the State Space Model (SSM), GLMamba establishes a unified global–local modeling framework. In the encoding stage, the network extracts global features via the state space structure and enhances multi-scale se mantic aggregation through the Change-aware Semantic Aggregation Module (CSAM). The Bi-temporal Fusion Attention (BFA) module further captures long-range cross-temporal dependencies and strengthens spatial detail representation. In the decoding stage, the Cross-Space Interaction Enhancement Module (CSIE) employs cross-attention to improve feature discrimination and spatial consistency in change regions. Experiments on LEVIR CD, GZ-CD, and SYSU-CD demonstrate that GLMamba consistently outperforms state-of-the-art (SOTA) methods in change detection accuracy, boundary preservation, and robustness to complex scenarios, while maintaining low parameter count and computational cost. These results validate the effectiveness and superiority of the proposed method. The code is publicly available at https://github.com/PXN222/GLMamba.


  • Top Cited Papers

    An SAM Fine-Tuning Framework With Frequency-Domain Interactive LoRA for Remote Sensing Change Detection

    by Junqing Huang, Shucheng Ji, Yapeng Wang, Min Xia, Xiaochen Yuan

    Achieving high-accuracy remote sensing change detection (RSCD) algorithms requires high-quality semantic feature extraction from remote sensing images (RSIs). Due to its powerful general-purpose feature extraction capability, the Segment Anything Model (SAM) has found wide application across diverse fields. However, SAM may not be optimally suited for RSIs. To address this limitation, we propose a Frequency-domain Interactive LoRA Fine-tuning Architecture (FILFArch) to enhance the performance of SAM in RSCD tasks. Based on FILFArch, we then develop two task-specific algorithms, the FILFBCD for Binary Change Detection (BCD), and the FILFSCD for Semantic Change Detection (SCD). To enhance the capability of SAM in capturing bi-temporal RSIs feature relationship, the Bi-temporal Feature Interactive LoRA (BIF-LoRA) is designed with Siamese architecture. Within BIF-LoRA, Frequency-Domain Feature Interaction (FDFI) utilizes Fast Fourier Transform Block (FFTB) to fuse bi-temporal frequency-domain features. This enables cross-temporal frequency-domain interaction, effectively discriminating spatio-temporal feature differences. Additionally, we use a shared BCD Decoder to serves as the binary change detector for both FILFBCD and FILFSCD. The BCD Decoder first applies a Coarse Difference Feature Extraction (CDFE) to coarsely fuse deep semantic features, yielding a coarse-grained change feature map. Subsequently, a Frequency-Domain Feature Enhancement (FDFE) refines these abstract features to generate a fine-grained change map. In FILFSCD, FDFE is further utilized to recover semantic change information of each temporal RSIs. Experimental results demonstrate that FILFBCD achieves the highest F1 scores of 83.53%, 66.75%, and 83.79% on BCD datasets MLCD, S2Looking, and SYSU-CD, respectively. Meanwhile, FILFSCD achieves the highest F1 scores of 64.05% and 87.02% on SCD datasets SECOND, and DSCD, respectively. These results demonstrate the effectiveness and versatility of the proposed FILFArch for RSCD tasks. The code is available at https://github.com/juncyan/filora.


  • A Triple-Branch Model With Style-Unified Contrastive Learning and Adaptive Dice Focal Loss for Change Detection

    by Zikai Zhao, Xingyu Liang, Min Xia, Liguo Weng, Haifeng Lin, Junqing Huang

    Deep learning has significantly reshaped the landscape of remote sensing change detection (RSCD). However, detecting subtle changes remains a formidable challenge due to the simultaneous presence of three critical issues: domain shift from seasonal and illumination variations, severe class imbalance between change and background areas, and the need for high deployment efficiency. To address these multifaceted problems in a unified manner, this paper introduces SDTNet, a comprehensive framework featuring three key innovations. First, to counter domain shift, we propose the Style-Unified Temporal Difference Contrastive Learning Strategy (STDCL), a novel, fine-grained contrastive learning method that guides the model to decouple real from pseudo changes without incurring additional inference costs. Second, to mitigate severe class imbalance, we design the Adaptive Dice Focal Loss (ADFLoss), which, unlike static loss functions, introduces two novel adaptive factors to dynamically balance sample weights and implement a curriculum learning strategy. Third, for efficient feature enhancement, we present the Dual-Branch Temporal Difference Transformer (DTDFormer), an efficiency-aware architecture that adapts the differential attention mechanism for bi-temporal inputs and incorporates a sparse attention module to reduce computational complexity. Experimental results demonstrate that our integrated SDTNet framework achieves state-of-the-art (SOTA) performance across five large-scale change detection datasets. The code is publicly available at https://github.com/Baobaon/SDTNet.

  • Open Access

    SAM-Based Efficient Feature Integration Network for Remote Sensing Change Detection: A Case Study on Macao Sea Reclamation

    by Junqing Huang, Junqi Bao, Min Xia, Xiaochen Yuan

    Sea reclamation expands land resources along coastal areas and supports sustainable urban growth. Sea reclamation monitoring is a significant application of remote sensing change detection (RSCD). Segment Anything Model (SAM) supports zero-shot image segmentation, but lacks general features of remote sensing images (RSIs), making it difficult to directly apply to RSCD. To utilize the visual recognition capabilities of SAM for improving RSCD, we propose an SAM-based Efficient Feature Integration Network (EFI-SAM). Random Fourier Features Adaptor (RFFA) is utilized to enhance Mobile SAM's capability for extracting general features from complex RSIs, resulting in Adaptive Mobile SAM, which serves as the feature extractor of EFI-SAM. To effectively integrate remote sensing difference features, Long-range Spatial Features Integration Module (LSFI) is designed to fuse the long-range spatial correlation of bi-temporal semantic features. Cross-dimensional Contextual Information Aggregation Decoder (CCIA) is then designed to effectively aggregate the multi-dimensional change information. Furthermore, we collected Macao RSIs through GEE to create a dataset, named Macao Land Change Detection (MLCD) dataset, which contains 10,000 images of size $256 \times 256$ . Experiments indicate that EFI-SAM achieves the highest metrics, with F1 of 80.75%, 75.86%, 82.86%, and 64.74% in MLCD, CLCD, SYSU-CD, and S2Looking, respectively. In addition to its high accuracy, it is noteworthy that EFI-SAM is a lightweight network with only 5.83M parameter. The code will be available at https://github.com/juncyan/efi-sam.git.


  • Top Cited Papers

    Interactive and Supervised Dual-Mode Attention Network for Remote Sensing Image Change Detection

    by Hongjin Ren, Min Xia, Liguo Weng, Haifeng Lin, Junqing Huang, Kai Hu

    With the rapid advancement of remote sensing technology, change detection using bitemporal remote sensing images has significant applications in land use planning and environmental monitoring. The emergence of Convolutional Neural Networks (CNNs) has accelerated the development of deep learning-based change detection. However, existing deep learning algorithms exhibit limitations in understanding bitemporal feature relationships and accurately identifying change region boundaries. Moreover, they inadequately explore feature interactions between bitemporal images before extracting differential features. To address these issues, this paper proposes a novel Interactive and Supervised Dual-Mode Attention Network (ISDANet). In the feature encoding stage, we employ the lightweight MobileNetV2 as the backbone to extract bitemporal features. Additionally, we design the Neighbor Feature Aggregation Module (NFAM) to aggregate semantic features from adjacent scales within the dual-branch backbone, enhancing the representation of temporal features. We further introduce the Interactive Attention Enhancement Module (IAEM), which effectively integrates self-attention and cross-attention mechanisms. This establishes deep interactions between bitemporal features, suppresses irrelevant noise, and ensures precise focus on true change regions. In the feature decoding stage, the Supervised Attention Module (SAM) reweights differential features and leverages supervisory signals to guide the learning of attention mechanisms, significantly improving boundary detection accuracy. SAM dynamically aggregates multi-level features, balancing high-level semantics and low-level details to capture subtle changes in complex scenes. The proposed model achieves F1 scores that are 0.28%, 1.6%, and 0.76% higher than the best comparative method, SEIFNet, on three CD datasets (LEVIR-CD, GZ-CD, and SYSU-CD), respectively, while maintaining a lightweight design with only 6.93M parameters and 3.46G FLOPs. The code is available at https://github.com/RenHongjin6/ISDANet.


  • Open Access

    LCCDMamba: Visual State Space Model for Land Cover Change Detection of VHR Remote Sensing Images

    by Junqing Huang, Xiaochen Yuan, Chan-Tong Lam, Yapeng Wang, Min Xia

    Land cover change detection (LCCD) is a crucial research topic for rational planning of land use and facilitation of sustainable land resource growth. However, due to the complexity of LCCD tasks, integrating global and local features and fusing contextual information from remote sensing features are essential. Recently, with the advent of Mamba, which maintains linear time complexity and high efficiency in processing long-range data, it offers a new solution to address feature-fusion challenges in LCCD. Therefore, a novel Visual State Space Model (SSM) for Land Cover Change Detection (LCCDMamba) is proposed, which uses Siam-VMamba as a backbone to extract multidimensional land cover features. To fuse the change information across difference temporal, Multi-scale Information Spatio-temporal Fusion (MISF) module is designed to aggregate difference information from bi-temporal features. The proposed MISF comprises Multi-Scale Feature Aggregation (MSFA) which utilizes strip convolution to aggregate multi-scale local change information of bi-temporal land cover features, and Residual with SS2D (RSS) which employs residual structure with SS2D to capture global feature differences of bi-temporal land cover features. To enhance the correlation of change features across different dimensions, in the Decoder, we design a Dual Token Modeling SSM (DTMS) through two token modeling approaches. This preserves high-dimensional semantic features and thus ensures that the multi-scale change information across various dimensions will not be lost during feature restoration. Experiments have been conducted on WHU-CD, LEVIR-CD, and GVLM datasets, and the results demonstrate that LCCDMamba achieves F1 scores of 94.18%, 91.68%, and 87.14%, respectively, outperforming all the models compared.


  • Top Cited Papers
    Open Access

    FTA-Net: Frequency-Temporal-Aware Network for Remote Sensing Change Detection

    by Taojun Zhu, Zikai Zhao, Min Xia, Junqing Huang, Liguo Weng, Kai Hu, Haifeng Lin, Wenyu Zhao

    Change detection (CD) aims to explore surface changes in co-aligned image pairs. However, many existing networks primarily focus on learning deep features, without considering the impact of attention and fusion strategies on detection performance. Therefore, a new Frequency-Temporal-Aware Network (FTA-Net) is proposed, it recognizes changes by means of a frequency-domain temporal fusion module and supervised attention to multilevel time-difference features, while reducing the model size. Frequency Temporal Fusion Module is designed to introduce the frequency attention mechanism into the fusion process. First, it has a two-branch Transformer-INN feature extractor using a Lite-Transformer that utilizes remote attention for low-frequency global features, and a Invertible Neural Network that focuses on extracting high-frequency local information. The semantic information and details of the object in both highfrequency and low-frequency feature maps are further strengthened by fusing the high-frequency local features and low-frequency global representations. Then, a Stepwise Modification Detection Module is proposed to better extract temporal difference information from bitemporal features. In addition, a Supervised Learning Module is constructed to re-weight features to efficiently aggregate multi-level features from highlevel to low-level. FTA-Net outperforms state-of-theart methods on three challenging CD datasets, and it have fewer parameters (4.93 M) and lower computational cost (6.71 G). Our code is available at https://github.com/Ztjdsb/FTA-Net