3 February 2023 in CAAI Transactions on Intelligence Technology
Even with an unprecedented breakthrough of deep learning in electroencephalography (EEG), collecting adequate labelled samples is a critical problem due to laborious and time‐consuming labelling. Recent study proposed to solve the limited label problem via domain adaptation methods. However, they mainly focus on reducing domain discrepancy without considering task‐specific decision boundaries, which may lead to feature distribution overmatching and therefore make it hard to match within a large domain gap completely. A novel self‐training maximum classifier discrepancy method for EEG classification is proposed in this study. The proposed approach detects samples from a new subject beyond the support of the existing source subjects by maximising the discrepancies between two classifiers' outputs. Besides, a self‐training method that uses unlabelled test data to fully use knowledge from the new subject and further reduce the domain gap is proposed. Finally, a 3D Cube that incorporates the spatial and frequency information of the EEG data to create input features of a Convolutional Neural Network (CNN) is constructed. Extensive experiments on SEED and SEED‐IV are conducted. The experimental evaluations exhibit that the proposed method can effectively deal with domain transfer problems and achieve better performance.
17 January 2025 in Pattern Recognition
14 November 2022 in arXiv
Crowd counting research has made significant advancements in real-world applications, but it remains a formidable challenge in cross-modal settings. Most existing methods rely solely on the optical features of RGB images, ignoring the feasibility of other modalities such as thermal and depth images. The inherently significant differences between the different modalities and the diversity of design choices for model architectures make cross-modal crowd counting more challenging. In this paper, we propose Cross-modal Spatio-Channel Attention (CSCA) blocks, which can be easily integrated into any modality-specific architecture. The CSCA blocks first spatially capture global functional correlations among multi-modality with less overhead through spatial-wise cross-modal attention. Cross-modal features with spatial attention are subsequently refined through adaptive channel-wise feature aggregation. In our experiments, the proposed block consistently shows significant performance improvement across various backbone networks, resulting in state-of-the-art results in RGB-T and RGB-D crowd counting.