Abstract:In order to improve the applicability of inattentive driving recognition method and the interpretability of recognition model, transfer learning was used to construct a drivers’ inattentive driving behavior recognition model, and the neural network visualization technology was applied to explain the model. Based on the VGG-16 model, the full connected layer of the original model is modified to adapt to the task of inattentive driving behavior recognition, ten kinds of driving behaviors in the original dataset were combined into six types according to the distraction types involved in the behavior, the combined dataset was used to train and verify the proposed model. Using Grad-Cam method, the focus areas of the model in identifying different driving behaviors were extracted and visualized. The characteristics of inattentive driving behaviors and the focused areas of the model for classification were compared to explain the model. The results show that the average accuracy of the model is 98.89% in the test set, and the trained model has the ability to locate the key features of each driving behavior and distinguish the behavior categories accordingly.