A Pilot Fatigue Detection Method Based on the YOLOv8 Model

Pilot action and behavior detection are one of the important means to evaluate pilots' workload. A detection method with high accuracy and low latency is a key challenge in current pilot fatigue driving detection. This paper proposes an improved YOLOv8 model-based method for detecting typical pilot actions and behaviors, aiming to achieve efficient and accurate detection of pilots' fatigue driving states. First, this study introduces the Coordinate Attention (CA) mechanism to enhance the model, establishing an improved YOLOv8 detection model that realizes the extraction of key features of typical actions and behaviors. Second, a lightweight detection head optimization strategy based on Partial Convolution (PConv) is adopted, addressing the issues of lightweight design and generalization ability of the improved model. Finally, this study constructs a dataset of typical actions and behaviors during the flight climb phase and conducts simulation experiments for training and testing. The experimental results show that the improved model achieves an accuracy of 97.7% in pilot fatigue feature detection, with the detection latency of 1.3ms, meeting the requirements of accuracy and real-time performance. This provides reliable technical support for pilot fatigue driving detection.