LDGCN: An Edge-End Lightweight Dual GCN Based on Single-Channel EEG for Driver Drowsiness Monitoring

TL;DR

LDGCN tackles real-time driver drowsiness monitoring from a single-channel EEG on edge devices. It introduces an interpretable Baseline Drowsiness Status Adjacency Graph (BDSAG), an Augmented Graph-level Module (AGM), and Adaptive Pruning Optimization (APO) to balance accuracy and resource usage, achieving efficient edge-end deployment. The approach leverages wavelet-based time-frequency features and a dual graph convolutional architecture with depthwise separable convolutions, while APO reduces inference latency by pruning channels and neurons. Experimental results on a single-eeg benchmark show state-of-the-art accuracy and significantly lower resource demand, with practical deployment on a 4th-generation Raspberry Pi, suggesting strong applicability for private, low-latency driver safety monitoring.

Abstract

Driver drowsiness electroencephalography (EEG) signal monitoring can timely alert drivers of their drowsiness status, thereby reducing the probability of traffic accidents. Graph convolutional networks (GCNs) have shown significant advancements in processing the non-stationary, time-varying, and non-Euclidean nature of EEG signals. However, the existing single-channel EEG adjacency graph construction process lacks interpretability, which hinders the ability of GCNs to effectively extract adjacency graph features, thus affecting the performance of drowsiness monitoring. To address this issue, we propose an edge-end lightweight dual graph convolutional network (LDGCN). Specifically, we are the first to incorporate neurophysiological knowledge to design a Baseline Drowsiness Status Adjacency Graph (BDSAG), which characterizes driver drowsiness status. Additionally, to express more features within limited EEG data, we introduce the Augmented Graph-level Module (AGM). This module captures global and local information at the graph level, ensuring that BDSAG features remain intact while enhancing effective feature expression capability. Furthermore, to deploy our method on the fourth-generation Raspberry Pi, we utilize Adaptive Pruning Optimization (APO) on both channels and neurons, reducing inference latency by almost half. Experiments on benchmark datasets demonstrate that LDGCN offers the best trade-off between monitoring performance and hardware resource utilization compared to existing state-of-the-art algorithms. All our source code can be found at https://github.com/BryantDom/Driver-Drowsiness-Monitoring.

Figures (8)

• Figure 1: The illustration of our motivation. It first use single-channel EEG (Oz) to construct the adjacency graph for representing its non-Euclidean space information and then use GCN deployed on edge device to identify the driver drowsiness. However, due to existing methods lack of consideration of neurophysiological knowledge on the characteristics of driver drowsiness, it reduces their interpretability and rationality in constructing adjacency graph, which directly affects the recognition accuracy.
• Figure 2: The overview of our drowsiness EEG signal monitoring.
• Figure 3: Schematic representation of ${\mathrm{x}_1}$ and BDST computation process. (a) Time complexity is $O(n^2)$ without adjacency connectivity coefficient K. (b) adjacency connectivity coefficient K = 8, the time complexity is $O(n)$.
• Figure 4: The adaptive channel-neuron pruning optimization of LDGCN.
• Figure 5: Framework diagram for pruned model deployment edge-end.