Robust sound target detection based on encoding and decoding models between sound and EEG signals

Sound target detection (STD) has received increasing attention in research areas such as industrial monitoring, medical diagnosis, and military detection. However, existing STD models face significant challenges in achieving robustness and generalization. In this study, to address the challenges, we develop an autonomous STD model that combines an encoding model, which maps sound signals to electroencephalogram (EEG) signals, with a decoding model, which identifies target information from EEG signals. Furthermore, we propose a model optimization strategy based on a weight loss function to balance the neural interpretability with sound recognition performance to enhance detection accuracy. Experiment results show that the proposed model demonstrates higher robustness and generalization, especially under low SNR conditions, than baseline methods. This work is the first to develop an autonomous STD model using brain-computer interfaces (BCIs) and consider auditory cognitive function in biomimetic sound detection, significantly advancing the application of BCI-based detection in real-world scenarios. Furthermore, it provides new insights into utilizing human intelligence through BCIs and offers new solutions for integrating BCIs with artificial intelligence technologies.