Talk · International symposium on image and signal processing and analysis/ISPA ... · October 29, 2025
End-to-End EEG Emotion Recognition via Adaptive Frequency Filtering and Sequential Denoising
Asfak Ali, Jotiraditya Banerjee, Vyacheslav V. Gulvanskii, Dmitrii Kaplun, Ram Sarkar
Abstract
In this paper, we address the challenge of noisy electroencephalogram (EEG) signals in emotion recognition by introducing a unified, end-to-end denoising and classification framework that operates across both frequency and time domains. First, we design an Adaptive Frequency Filter (AFF) module that learns to suppress spectral noise while preserving emotion-relevant frequency components; this filtering is performed as an integral layer within the neural architecture, ensuring seamless gradient-based optimization. To further refine temporal signal integrity, we incorporate a two-layered Long Short-Term Memory (LSTM) denoiser: the first LSTM layer captures coarse temporal dependencies and removes bulk noise, and the second layer targets finer residual artifacts, thereby enhancing the sequential features fed to the classifier. To guide the network toward both classification accuracy and parsimonious frequency utilization, we propose a novel Multi-Objective Learning Loss (MOLL), which combines label-smoothed categorical cross-entropy with an$L_{1}$- based regularization on the learned frequency mask-encouraging sparsity and robustness across emotional states. We validated our approach on the widely used WESAD, SWELL and SEED emotion datasets, performing better in comparison with state-of-the-art baselines. We tested our model on different devices and found that it can run in real-time, with inference times as low as 22 ms on GPUs and under 130 ms on edge devices like Raspberry Pi 5, making it suitable for real-world emotion recognition applications. The code materials for the following work will be available at: https://github.com/praetorian2710/EEG-Emotion-Recognition-via-Adaptive-Frequency-Filtering-and-Sequential-Denoising
More by Dmitry Kaplun
- Codewithsayanjib/slm-systematic-review: v1.1Zenodo · 2026
- Microscopy Image Dataset of pulmonary vessels for Quantitative assessment of fibrosisScientific Data · 2026
- Microscopy Image Dataset of pulmonary vessels for Deep Learning-Based Quantitative assessment of fibrosisFigshare · 2026
- MA2MU-Net++: A Mahalanobis distance and Ackley function based multi-modal U-Net++ model for breast nuclei segmentation in histopathology imagesBiomedical Signal Processing and Control · 2026