10 July 2025 in IEEE Sensors Journal
Fusion of complementary feature information from visible-infrared images can effectively improve the performance of target detection. However, the information of visible images is obscured in low-light environments, resulting in the unsatisfactory performance of existing visible-infrared target detection methods in low-light environments. To address this problem, we introduce dual-modality fusion detection with image-enhancement and four-head adaptively spatial feature fusion network (DFIF), which is to improve the multi-target detection accuracy in low-light environments. DFIF is able to mine more information about the target during the cross-modality fusion process through the proposed low light image enhancement network (LLIE). Furthermore, considering the differences and complementarities between the two modalities, we design a dual-modality attention feature fusion module in the dual-stream feature extraction process to fully integrate the complementary information in the feature extraction stage. Finally, we design four-head adaptively spatial feature fusion network (FASFF) with a big target detection layer that can adaptively learn spatial weights for feature mapping fusion at different scales. By optimizing the weighting parameters, FASFF effectively integrates the features at different levels to achieve higher accuracy target detection. We evaluate it on the LLVIP and M3FD datasets, and our DFIF achieves 74.1% and 73.9% on the mAP metrics, respectively, demonstrating optimal detection performance and establishing a new benchmark for subsequent dual-modality target detection.
7 May 2026 in IEEE Transactions on Affective Computing
Recent studies have demonstrated the effectiveness of Graph Neural Networks (GNNs) in EEG-based depression detection. However, two critical challenges remain unresolved. These challenges limit the ability of GNN-based methods to fully exploit spatio-temporal (ST) features for EEG-based depression detection. Firstly, the depression-relevant cross-temporal interactions that represent the complex relationships between different electrodes across various time steps are not fully considered. Secondly, although depression is closely associated with multiple distributed brain regions and their interactions, the existing methods provide limited exploration of brain-region-level ST features, failing to fully exploit the depression-related neural patterns at the brain-region level. To tackle these limitations, in this paper we propose a Dual-branch Cross-Temporal Graph Neural Network (DCTGNN) to effectively capture cross-temporal (CT) interactions. Specifically, a CT graph structure is introduced to explicitly model relationships between electrodes (or regions) across different time steps. To extract region-level ST features, we design a dedicated region-level branch that integrates electrode features into brain regions. A region-level CT graph is then constructed to capture region-level ST interactions. Extensive experiments validate the effectiveness of DCTGNN, achieving consistent improvements across the three datasets. Furthermore, CT graph-based analysis reveals that the frontal and parietal lobes are highly associated with depression, aligning with previous studies.
17 March 2026 in IEEE Transactions on Computational Social Systems
P300-based brain–computer interfaces (BCIs) enable direct communication between the brain and external devices by decoding P300 potentials, playing a vital role in rehabilitation and cognitive neuroscience research. Accurate detection of P300 potentials is essential for the successful implementation of P300-based BCIs. Current mainstream P300 detection algorithms are based on multichannel electroencephalogram (EEG) signals, and while promising detection results have been achieved, they still suffer from the following issues: 1) insufficient exploitation of the distinct and characteristic overall temporal trend of P300 potentials; and 2) oversimplified aggregation of multichannel EEG information without effectively utilizing the non-Euclidean topological relationships between EEG channels. To address the above issues, we propose a trend-aware multiscale spatial-temporal graph convolutional neural network (TMSGCN) for P300 detection. Specifically, to effectively capture the long-term temporal trend of P300 potentials to improve detection robustness, we explicitly extract the trend component from raw EEG signals along time dimension and utilize it to assist the identification of P300 potentials. Subsequently, a multiscale temporal convolution module (MTCM) is applied to extract multitime scale amplitude features from the processed EEG signals, serving as input for an adaptive graph convolution module (AGCM). The AGCM effectively models the intricate inter-channel relationships as a graph by complementarily considering the structural and functional connectivity of human brain, thereby capturing more discriminative and informative spatial features related to P300 potentials. Extensive experiments on three datasets demonstrate the superiority of TMSGCN over other state-of-the-art P300 detection methods. Furthermore, the results of ablation study and visualization experiments indicate the effectiveness of each component in TMSGCN.