Next Generation Imaging in Consumer Technology for ERP Detection-Based EEG Cross-Subject Visual Object Recognition

• Published in: IEEE transactions on consumer electronics Vol. 70; no. 1; pp. 3688 - 3696
• Main Authors: Bhatt, Mohammed Wasim, Sharma, Sparsh
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.02.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial neural networks; Brain; Brain modeling; Consumer electronics; consumer technology; Convolution; Convolutional neural networks; cross-subject analysis; Electroencephalography; Event-related potentials; Feature extraction; graph representation; Graph representations; Graphical representations; Man-machine interfaces; Object recognition; Spatial data; visual object recognition; Visualization
• ISSN: 0098-3063; 1558-4127
• DOI: 10.1109/TCE.2024.3368569

The perception and recognition of objects are essential for meeting consumer needs in the realm of consumer technology. Current research exploring the association between variations in brain activity and their prospective application in user-friendly brain-machine interfaces (BMIs) has been growing significant momentum. To this end, a novel model is proposing that enhance the detection of event-related potentials (ERP) from EEG signals, particularly for visual object recognition across different subjects, incorporating next generation imaging technology tailored for consumer electronics. It utilizes a graph representation that captures EEG spatial information, across all subjects and tasks. It merges a convolutional neural network (CNN) with a long short-term memory network (LSTM), creating a solid CNN-LSTM architecture followed by dual-attention mechanism. It includes both selective kernel convolution and self-attention mechanisms. They jointly work to precisely capture the unique spatiotemporal characteristics of EEG signals from various subjects. This results in boosting the accuracy of ERP detection for individuals. Experimental validation of the proposed model shows promising results. It was tested on a comprehensive benchmark dataset designed around the rapid serial visual presentation paradigm. The data shows that this new method outperforms seven existing ERP detection techniques in scenarios involving different subjects.