Tao, S., Hu, J., Goh, W. L., & Gao, Y. (2024). Squeeze-Excite Fusion Based Multimodal Neural Network for Sleep Stage Classification with Flexible EEG/ECG Signal Acquisition Circuit. 2024 IEEE International Symposium on Circuits and Systems (ISCAS), 1–5. https://doi.org/10.1109/iscas58744.2024.10557984
Abstract:
This paper presents a multimodal fusion strategy for sleep stage classification based on polysomnography (PSG) which consists of electroencephalogram (EEG) and Electrocardiogram (ECG) data. By implementing the Squeeze-Excite (SE) Fusion mechanism, a more effective contribution of EEG and ECG signals to the neural network classification is ensured. To address the challenges of imbalance in the dataset, a balanced sampler is introduced. The EEG signals are transposed to the frequency domain utilizing linear-frequency cepstrum coefficients (LFCC), enhancing the feature extraction process. A recurrent convolutional neural network (RCNN) is employed to reduce the number of model parameters and further optimize the architecture. The weight of the network is then quantized down to INT4 to ensure hardware compatibility, particularly for edge devices. By employing these methods on signals, this optimized approach has achieved a validation accuracy of 77.6% with only 23.5KB weight memory size on the MIT-BIH dataset across six classification categories.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the Agency for Science, Technology and Research (A*STAR), Singapore - Nanosystems at the Edge Programme
Grant Reference no. : A18A1b0055