Название: Brain Fingerprint Identification
Автор: Wanzeng Kong, Xuanyu Jin
Издательство: Springer
Серия: Brain Informatics and Health
Год: 2025
Страниц: 202
Язык: английский
Формат: pdf (true), epub
Размер: 41.8 MB

This book delves into the emerging field of biometric identification using brainwave patterns. Specifically, this book presents recent advances in electroencephalography (EEG)-based biometric recognition to identify unique neural signatures that can be used for secure authentication and identification.

Traditional biometric systems such as fingerprints, iris scans, and face recognition have become integral to security and identification. However, these methods are increasingly vulnerable to spoofing and other forms of attack. Unlike other traditional biometrics, EEG signals are non-invasive, continuous authentication, liveness detection, and resistance to coercion due to the complexity and uniqueness of brain patterns. Therefore, it is particularly suitable for high-security fields such as military and finance, providing a promising alternative for future high-security identification and authentication.

However, most of the existing brain fingerprint identification studies require subjects to perform specific cognitive tasks, which limits the popularization and application of brain fingerprint identification in practical scenarios. Additionally, due to the low signal-to-noise ratio (SNR) and time-varying characteristics of EEG signals, there are distribution differences in EEG data across sessions from several days, leading to stability issues in brain fingerprint features extracted at different sessions. Finally, because the EEG signal is affected by the coupling of multiple factors and the nervous system has continuous spontaneous variability, which makes it difficult for the brain fingerprint identification model to be suitable for the scenarios of unseen sessions and cognitive tasks, and there is the problem of insufficient model generalization. In this book, based on traditional Machine Learning methods and Deep Learning methods, the authors will carry out multi-task single-session, single-task multi-session, and multi-task multi-session brain fingerprint identification research respectively for the above problems, to provide an effective solution for the application of brain fingerprint identification in practical scenarios.

Deep Learning has demonstrated a remarkable ability to extract high-level features and uncover complex latent dependencies, making it highly effective for various tasks. However, the success of Deep Learning models typically hinges on the availability of large datasets for training, which poses a significant challenge in real-world applications. In the domain of brain fingerprint identification, where data from multiple individuals is often sparse and each class contains only a few samples, Deep Learning models face difficulties in achieving reliable performance. This chapter presents a Convolutional Tensor-Train Neural Network (CTNN) designed to tackle these challenges in the context of multi-task brain fingerprint identification with limited training samples. The method integrates a convolutional neural network (CNN) with a depthwise separable convolution mechanism to extract local temporal and spatial features from the brainprint, focusing on subtle neural patterns that distinguish individuals. To further enhance the model’s ability to capture complex interdependencies, the TensorNet (TN) component is introduced, employing low-rank tensor decomposition to model multilinear interactions between different features. This representation enables the model to efficiently integrate local information into a global feature space with minimal parameters, making it particularly effective in scenarios with small-sample sizes. The CTNN approach not only addresses the challenge of limited data but also excels in multi-task learning, enabling it to extract shared features across different recognition tasks. This capability is crucial for real-world applications where brainprint identification must function across diverse individuals and conditions. Furthermore, the model provides interpretability by identifying key brain regions, with seven specific channels being dominant in the recognition tasks, thus offering valuable insights into the neural biomarkers underlying brain fingerprint identification.