RESEARCH ON NON-INVASIVE DROWSINESS DETECTION METHOD FOR HARVESTER DRIVERS BASED ON MULTI-FEATURE FUSION

Driver drowsiness severely impairs the normal operation of harvesters, leading to casualties and economic losses. Effectively detecting driver drowsiness in harvesters remains a significant challenge. This paper introduces a lightweight convolutional neural network that identifies driver drowsiness in harvester operators by analyzing the driver's eyes, mouth, and head posture. The model comprises a lightweight CNN, a Long Short-Term Memory (LSTM) network, and an attention layer, achieving high efficiency and low latency. Experimental results demonstrate that the CNN-LSTM-Attention model effectively balances accuracy and computational efficiency, enabling rapid and precise drowsiness detection. This approach significantly improves safety during combine harvester operation.