A Combination Perception Model Based on CNN-BiGRU-Attention and Its Application in Signal Processing

Deep learning-based long and short-term memory networks offer notable advantages in processing datasets that vary over time. Current models, however, face challenges such as poor information capture and overfitting/underfitting, especially with non-periodic patterns in long signals. To address these issues, this paper proposes an enhanced CNN-BiGRU-Attention model. This model integrates a convolutional neural network (CNN) for extracting spatial features, a bidirectional gated recurrent unit (BiGRU) for handling long-term dependencies, and an attention mechanism to highlight key information dynamically. The attention mechanism improves the model's focus on critical features within long sequences. Tested on the Sleep-EDF dataset, the model achieved an accuracy (ACC) of 73.25% and an F1 score of 72.24% on the training set. Compared to existing models, the proposed model shows a 5% increase in ACC and a 4% increase in F1 score, demonstrating its effectiveness in capturing complex patterns and focusing on essential details.