Driver fatigue driving state identification method based on deep learning

Abstract

The invention discloses a driver fatigue driving state recognition method based on deep learning, and relates to the technical field of computer vision recognition methods, and the method comprises the following steps: S01, multi-modal data collection, S02, data preprocessing, S03, data enhancement, S04, multi-modal feature fusion, S05, deep learning model construction, S06, fatigue driving state recognition, and S07, real-time feedback and optimization. S08, a fatigue driving state recognition result is output, the fatigue characteristics of the driver can be captured from multiple dimensions by fusing multi-modal data including video images, vehicle driving data, environment data, voice signals and physiological signals, compared with the prior art, the scheme can more comprehensively analyze the fatigue state of the driver, and the fatigue driving state recognition result is obtained. Therefore, traffic accidents caused by fatigue driving are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision recognition methods, and specifically to a method for recognizing a driver's fatigue driving state based on deep learning. Background Art

[0002] With the increase in the number of cars and the rapid development of road transportation, traffic accidents caused by fatigue driving are increasing, posing a serious threat to people's lives and property safety. After searching, the Chinese patent publication number CN118196766A discloses a method for identifying driver fatigue driving behavior based on deep learning. Although this method is based on the recognition algorithm research of fatigue driving behavior under deep learning, on the basis of pre-processing the video image of the driver's fatigue driving behavior to improve the image quality, after extracting the behavior features, a fatigue driving behavior recognition model is constructed based on the deep learning framework, which is used as input to realize fatigue driving behavior recognition, thereby reducing the occurrence of traffic accidents; however, the prior art has the following deficiencies: Dependence on a single data source: Existing technologies mostly rely on a single data source, such as video images or vehicle driving data, and lack fusion analysis of multimodal data, resulting in limited recognition accuracy; Poor environmental adaptability: Existing methods have poor adaptability to different lighting and weather conditions, which affects the recognition effect; Lack of dynamic adjustment mechanism: Existing fatigue driving recognition methods lack the ability to dynamically adjust recognition strategies based on individual differences of drivers and real-time data; To this end, the present invention provides an improved method for identifying driver fatigue driving status based on deep learning; by fusing multimodal data and utilizing advanced deep learning architecture and algorithms, accurate identification of the driver's fatigue status is achieved. Summary of the invention

[0003] In view of the shortcomings of the prior art, the present invention provides a method for identifying driver fatigue driving status based on deep learning, which solves the problems of the existing methods proposed in the background technology, such as dependence on a single data source, poor environmental adaptability and lack of a dynamic adjustment mechanism.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for identifying driver fatigue driving status based on deep learning, the steps of which are as follows: S01: Multimodal data collection, using multi-angle high-definition cameras to collect video images of the driver's face, eyes and head; collecting vehicle driving data, including vehicle speed, steering wheel angle, braking frequency; collecting vehicle internal environment data, such as temperature, humidity, and light intensity; collecting the driver's voice signal through the microphone in the car; installing millimeter wave radar to monitor the driver's breathing rate and heartbeat signal in real time; S02: Data preprocessing, enhancing the collected video images, including illumination correction, contrast enhancement and noise filtering; normalizing the vehicle driving data; preprocessing the speech signal to extract emotional features; filtering the respiratory rate and heartbeat signals to extract physiological features related to fatigue; S03: Data enhancement: Use the spatiotemporal graph convolutional network (ST-GCN) to enhance the driver behavior data in the video image and capture the spatiotemporal relationship in the video sequence; introduce quantum deep learning to efficiently extract features and enhance multimodal data; use the generative adversarial network (GAN) to enhance the environmental perception data and generate realistic virtual environment data; S04: Multimodal feature fusion: fuse the pre-processed video image features, vehicle driving data features, environmental data features, voice emotion features and physiological signal features to construct a multimodal feature fusion vector; use the attention mechanism to dynamically assign weights to different modal data; introduce a multimodal heterogeneous data fusion framework, and build a heterogeneous graph neural network (HGN) to model the complex relationship between different modal data as a graph structure; S05: Deep learning model construction: construct a deep learning model based on the Transformer architecture, with the input being a multimodal feature fusion vector and the output being the driver’s fatigue state probability; introduce a meta-learning mechanism so that the model can quickly adapt to individual differences between different drivers and changes in the driving environment; introduce an adaptive dynamic threshold mechanism to dynamically adjust the fatigue state judgment threshold based on real-time data; introduce quantum deep learning to optimize the model’s training process; S06: Fatigue driving state recognition: input the pre-processed multimodal data into the constructed deep learning model. The model outputs the driver's fatigue state probability by learning the correlation relationship of multimodal features; judges whether the driver is in a fatigue state according to the set threshold, and issues an alarm signal in real time; S07: Real-time feedback and optimization, introduce reinforcement learning mechanism, adjust model parameters according to real-time feedback; introduce online learning mechanism, the model will update itself according to new data after each recognition, and continuously optimize the recognition effect; introduce quantum reinforcement learning, use the parallel processing capability of quantum computing to optimize the reinforcement learning process; S08: Outputting the fatigue driving state recognition result. Outputting the final fatigue driving state recognition result to the vehicle control system or the driver warning system so that corresponding safety measures can be taken.

[0005] Preferably, in step S01, the acquisition frequency of the multi-angle high-definition camera is 25-60 frames per second, and the resolution of the camera is not less than 1920×1080 pixels; the monitoring frequency of the millimeter wave radar is not less than 10 Hz.

[0006] Preferably, in step S02, the video image enhancement process includes illumination correction and contrast enhancement, wherein illumination correction adopts a histogram equalization algorithm, and contrast enhancement adopts an adaptive histogram equalization algorithm, and the specific formula is:

[0007] in, is the input image, and are the mean and standard deviation of the image, respectively. is the contrast enhancement factor, ranging from 1.5 to 3.0.

[0008] Preferably, in step S03, the convolution kernel order of the spatiotemporal graph convolutional network (ST-GCN) is 3, which can effectively capture the spatiotemporal relationship in the video sequence, and the feature extraction formula is:

[0009] in, is the input feature matrix, is the adjacency matrix, is the degree matrix, is the learnable weight parameter, is the activation function is the order of the convolution kernel; the quantum deep learning adopts the quantum convolutional neural network (QCNN), which uses the superposition and entanglement characteristics of quantum bits to significantly improve the efficiency and accuracy of feature extraction. The feature extraction formula is:

[0010] in, is the input quantum state, is the output quantum state, The adversarial network (GAN) performs quantum convolution operation on the environmental perception data to generate realistic virtual environment data; the generator and the discriminator Both use a deep convolutional neural network structure, where the output resolution of the generator is no less than 1920×1080 pixels to ensure the high quality and fidelity of the generated data; the loss functions of the generator and discriminator are defined as follows: The loss function of the generator for:

[0011] in, From the prior distribution The noise vector sampled in , is the generator according to the noise vector The data generated, is the discriminator; the loss function of the discriminator for:

[0012] in, From the real data distribution The data sampled in is the discriminator’s judgment result on the real data. is the discriminator’s judgment result on the generated data.

[0013] Preferably, in step S04, the multimodal heterogeneous data fusion framework adopts a heterogeneous graph neural network (HGN), and the node feature update adopts a graph attention network (GAT) architecture, and the specific formula is:

[0014] in, For the node The characteristics of the layer, is the set of neighbor nodes of the node, is the attention weight, is the learnable weight matrix, is the activation function.

[0015] Preferably, in step S05, the deep learning model based on the Transformer architecture adopts an encoder-decoder structure, with 6 encoder layers, 6 decoder layers, and 512 hidden units in each layer; the meta-learning mechanism adopts model-independent meta-learning (MAML), which can quickly adapt to new tasks and environments; the adaptive dynamic threshold mechanism dynamically adjusts the threshold according to the statistical characteristics of real-time data, and the threshold adjustment range is 0.1-0.9.

[0016] Preferably, in step S06, the fatigue state judgment threshold is dynamically adjusted according to the driver's historical data and real-time data, and the threshold adjustment step is 0.01; the alarm signal includes a visual alarm, an auditory alarm and a tactile alarm, the visual alarm uses a red flashing light, the auditory alarm uses a 85-90dB beep, and the tactile alarm uses seat vibration.

[0017] Preferably, in step S07, the reinforcement learning mechanism adopts a deep Q network (DQN), the learning rate α is 0.001, and the discount factor γ is 0.99; the online learning mechanism adopts an incremental learning algorithm, and the sample size for each update is 10-50; the quantum reinforcement learning adopts a quantum Q-learning algorithm, and the number of quantum bits is 5-10.

[0018] Preferably, in step S08, the fatigue driving state recognition result is output through the vehicle control system, and the vehicle control system takes corresponding safety measures according to the recognition result, including: when the recognition result is a fatigue state, the vehicle automatically decelerates to a safe speed, and the safe speed range is 20-60km / h; at the same time, the vehicle automatically turns on the automatic driving assistance function, including lane keeping assist and automatic braking assist.

[0019] The present invention provides a method for identifying driver fatigue driving status based on deep learning. It has the following beneficial effects: (1) Improving the accuracy of fatigue driving status recognition: By fusing multimodal data, including video images, vehicle driving data, environmental data, voice signals, and physiological signals, the driver's fatigue characteristics can be captured from multiple dimensions. In particular, the spatiotemporal graph convolutional network (ST-GCN) is used to capture the spatiotemporal relationship in the video sequence, and quantum deep learning is used to improve the efficiency and accuracy of feature extraction. The application of these technologies has greatly improved the accuracy of fatigue driving status recognition. Compared with the existing technology, this solution can analyze the driver's fatigue status more comprehensively, thereby reducing traffic accidents caused by fatigue driving. (2) Enhance the real-time and dynamic adaptability of the system: The present invention introduces an adaptive dynamic threshold mechanism and an online learning mechanism, so that the system can dynamically adjust the recognition strategy according to real-time data and individual differences of drivers. This dynamic adjustment capability enables the system to not only adapt to different driving environments and conditions, but also optimize the recognition effect in real time according to the driver's behavior and physiological changes, thereby significantly improving the real-time and adaptability of the system, solving the problem of the lack of dynamic adjustment of the recognition strategy according to individual differences and real-time data of drivers in the prior art, making the recognition of fatigue driving status more flexible and accurate; (3) Improve the generalization ability and robustness of the system: By using generative adversarial networks (GANs) to generate realistic virtual environment data, this solution can enhance the training data of the model and improve the model's recognition ability for unseen data. In addition, by introducing quantum reinforcement learning to optimize the reinforcement learning process and utilizing the parallel processing capability of quantum computing, the training efficiency and generalization ability of the model are further improved. The combined use of these technologies enables the present invention to maintain good recognition performance between different vehicles, different driving environments and different drivers, thereby improving the generalization ability and robustness of the entire system and solving the problem of poor adaptability of the prior art in different environments.

[0020] This solves the problems of existing methods such as reliance on a single data source, poor environmental adaptability, and lack of a dynamic adjustment mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] See also Figure 1 The embodiment of the present invention provides a technical solution: a method for identifying a driver's fatigue driving state based on deep learning, the steps of which are as follows: S01: Multimodal data collection, using multi-angle high-definition cameras to collect video images of the driver's face, eyes and head; collecting vehicle driving data, including vehicle speed, steering wheel angle, braking frequency; collecting vehicle internal environment data, such as temperature, humidity, and light intensity; collecting the driver's voice signal through the microphone in the car; installing millimeter wave radar to monitor the driver's breathing rate and heartbeat signal in real time; S02: Data preprocessing, enhancing the collected video images, including illumination correction, contrast enhancement and noise filtering; normalizing the vehicle driving data; preprocessing the speech signal to extract emotional features; filtering the respiratory rate and heartbeat signals to extract physiological features related to fatigue; S03: Data enhancement: Use the spatiotemporal graph convolutional network (ST-GCN) to enhance the driver behavior data in the video image and capture the spatiotemporal relationship in the video sequence; introduce quantum deep learning to efficiently extract features and enhance multimodal data; use the generative adversarial network (GAN) to enhance the environmental perception data and generate realistic virtual environment data; S04: Multimodal feature fusion: fuse the pre-processed video image features, vehicle driving data features, environmental data features, voice emotion features and physiological signal features to construct a multimodal feature fusion vector; use the attention mechanism to dynamically assign weights to different modal data; introduce a multimodal heterogeneous data fusion framework, and build a heterogeneous graph neural network (HGN) to model the complex relationship between different modal data as a graph structure; S05: Deep learning model construction: construct a deep learning model based on the Transformer architecture, with the input being a multimodal feature fusion vector and the output being the driver’s fatigue state probability; introduce a meta-learning mechanism so that the model can quickly adapt to individual differences between different drivers and changes in the driving environment; introduce an adaptive dynamic threshold mechanism to dynamically adjust the fatigue state judgment threshold based on real-time data; introduce quantum deep learning to optimize the model’s training process; S06: Fatigue driving state recognition: input the pre-processed multimodal data into the constructed deep learning model. The model outputs the driver's fatigue state probability by learning the correlation relationship of multimodal features; judges whether the driver is in a fatigue state according to the set threshold, and issues an alarm signal in real time; S07: Real-time feedback and optimization, introduce reinforcement learning mechanism, adjust model parameters according to real-time feedback; introduce online learning mechanism, the model will update itself according to new data after each recognition, and continuously optimize the recognition effect; introduce quantum reinforcement learning, use the parallel processing capability of quantum computing to optimize the reinforcement learning process; S08: Outputting the fatigue driving state recognition result. Outputting the final fatigue driving state recognition result to the vehicle control system or the driver warning system so that corresponding safety measures can be taken.

[0024] In step S01, the acquisition frequency of the multi-angle high-definition camera is 25-60 frames per second, and the resolution of the camera is not less than 1920×1080 pixels; the monitoring frequency of the millimeter wave radar is not less than 10 Hz.

[0025] In step S02, the video image enhancement process includes illumination correction and contrast enhancement, wherein illumination correction adopts a histogram equalization algorithm, and contrast enhancement adopts an adaptive histogram equalization algorithm, and the specific formula is:

[0026] in, is the input image, and are the mean and standard deviation of the image, respectively. is the contrast enhancement factor, ranging from 1.5 to 3.0.

[0027] In step S03, the convolution kernel order of the spatiotemporal graph convolutional network (ST-GCN) is 3, which can effectively capture the spatiotemporal relationship in the video sequence. The feature extraction formula is:

[0028] in, is the input feature matrix, is the adjacency matrix, is the degree matrix, is the learnable weight parameter, is the activation function is the order of the convolution kernel; quantum deep learning uses quantum convolutional neural network (QCNN), which uses the superposition and entanglement characteristics of quantum bits to significantly improve the efficiency and accuracy of feature extraction. The feature extraction formula is:

[0029] in, is the input quantum state, is the output quantum state, For quantum convolution operation, the adversarial network (GAN) enhances the environmental perception data to generate realistic virtual environment data; Generator and the discriminator Both use a deep convolutional neural network structure, where the output resolution of the generator is no less than 1920×1080 pixels to ensure the high quality and fidelity of the generated data; the loss functions of the generator and discriminator are defined as follows: The loss function of the generator for:

[0030] in, From the prior distribution The noise vector sampled in , is the generator according to the noise vector The data generated, is the discriminator; the loss function of the discriminator for:

[0031] in, From the real data distribution The data sampled in is the discriminator’s judgment result on the real data. is the discriminator’s judgment result on the generated data.

[0032] In step S04, the multimodal heterogeneous data fusion framework adopts a heterogeneous graph neural network (HGN), and the node feature update adopts a graph attention network (GAT) architecture. The specific formula is:

[0033] in, is the feature of the node at the layer, is the set of neighbor nodes of the node, is the attention weight, is the learnable weight matrix, is the activation function.

[0034] In step S05, the deep learning model based on the Transformer architecture adopts an encoder-decoder structure, with 6 encoder layers, 6 decoder layers, and 512 hidden units in each layer; the meta-learning mechanism adopts model-independent meta-learning (MAML), which can quickly adapt to new tasks and environments; the adaptive dynamic threshold mechanism dynamically adjusts the threshold according to the statistical characteristics of real-time data, and the threshold adjustment range is 0.1-0.9.

[0035] In step S06, the fatigue status judgment threshold is dynamically adjusted according to the driver's historical data and real-time data, and the threshold adjustment step is 0.01; the alarm signal includes visual alarm, auditory alarm and tactile alarm, the visual alarm uses red flashing light, the auditory alarm uses 85-90dB beeping sound, and the tactile alarm uses seat vibration.

[0036] In step S07, the reinforcement learning mechanism adopts the deep Q network (DQN), the learning rate α is 0.001, and the discount factor γ is 0.99; the online learning mechanism adopts the incremental learning algorithm, and the sample size for each update is 10-50; the quantum reinforcement learning adopts the quantum Q-learning algorithm, and the number of quantum bits is 5-10.

[0037] In step S08, the fatigue driving status recognition result is output through the vehicle control system, and the vehicle control system takes corresponding safety measures according to the recognition result, including: when the recognition result is a fatigue state, the vehicle automatically decelerates to a safe speed within a range of 20-60km / h; at the same time, the vehicle automatically turns on the automatic driving assistance function, including lane keeping assist and automatic braking assist.

[0038] In order to verify the effectiveness of the present invention, a series of experiments were designed and compared with the traditional method. The experimental data are shown in Table 1 and Table 2 Table 1: Comparison of fatigue driving status recognition performance Performance Indicators This program Traditional methods Improvement ratio (%) Accuracy (%) 92.0 85.5 7.6% Recall rate (%) 90.9 80.0 13.6% Accuracy(%) 91.8 82.3 11.0% F1 score 0.955 0.82 16.5% ROC AUC 0.98 0.75 26.7% Average response time (ms) 120 250 -52% CPU usage (%) 15 25 -40% Memory usage (MB) 200 300 -33.3% The table lists the comparison between the proposed solution and the traditional method in the recognition performance of fatigue driving status, including eight key performance indicators: accuracy, recall, precision, F1 score, area under the ROC curve (AUC), average response time, CPU usage, and memory consumption; the improvement ratio column shows the improvement percentage of the proposed solution compared with the traditional method in each performance indicator; This solution is significantly better than traditional methods in all key performance indicators, especially in accuracy and F1 score, with improvements of 11.0% and 16.5% respectively, demonstrating the superior performance of this solution in fatigue driving status identification.

[0039] Table 2: Comparison of fatigue driving status recognition performance 2 Performance Indicators This program Traditional methods Improvement ratio (%) Recognition speed (frames / second) 30 15 100% False alarm rate (%) 2.0 5.0 60% Multimodal data fusion accuracy (%) 95.0 88.0 7.95% Real-time processing capability (yes / no) yes no - Environmental adaptability score (1-10) 9.5 7.5 26.7% User satisfaction (1-5) 4.8 3.6 33.3% The table lists the comparison between this solution and the traditional method in the recognition performance of fatigue driving status, including six key performance indicators: recognition speed, false alarm rate, multimodal data fusion accuracy, real-time processing capability, environmental adaptability score and user satisfaction.

[0040] The improvement ratio column shows the improvement percentage of the proposed solution in each performance indicator compared with the traditional method.

[0041] This solution achieved a 100% improvement in recognition speed, indicating that it can process more video frames and provide a smoother recognition experience.

[0042] In terms of false alarm rate, this scheme achieved a 60% reduction, showing a significant improvement in reducing false alarms.

[0043] The improvement in the accuracy of multimodal data fusion demonstrates the efficiency of this scheme in integrating information from different data sources.

[0044] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0045] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for identifying driver fatigue driving status based on deep learning, characterized in that: The steps are as follows: S01: Multimodal data collection, using multi-angle high-definition cameras to collect driver's face, eyes and head video images; while collecting vehicle driving data; S02: Data preprocessing, enhancing the collected video images; S03: Data enhancement, using a spatiotemporal graph convolutional network to enhance the driver behavior data in the video image and capture the spatiotemporal relationship in the video sequence; Introducing quantum deep learning to efficiently extract features and enhance processing of multimodal data; S04: Multimodal feature fusion, fusing the pre-processed video image features, vehicle driving data features, environmental data features, voice emotion features and physiological signal features to construct a multimodal feature fusion vector; S05: Deep learning model construction: construct a deep learning model based on the Transformer architecture, with the input being a multimodal feature fusion vector and the output being the driver's fatigue state probability; introduce a meta-learning mechanism; S06: Fatigue driving status recognition: input the pre-processed multimodal data into the constructed deep learning model. The model learns the correlation between multimodal features and issues an alarm signal in real time. S07: Real-time feedback and optimization, introduce reinforcement learning mechanism, adjust model parameters according to real-time feedback; introduce online learning mechanism, the model will update itself according to new data after each recognition; S08: Outputting the fatigue driving state recognition result. Outputting the final fatigue driving state recognition result to the vehicle control system or the driver warning system so that corresponding safety measures can be taken.

2. The method for identifying driver fatigue driving status based on deep learning according to claim 1, characterized in that: In step S01, the acquisition frequency of the multi-angle high-definition camera is 25-60 frames per second, and the resolution of the camera is not less than 1920×1080 pixels; the monitoring frequency of the millimeter wave radar is not less than 10 Hz.

3. The method for identifying driver fatigue driving status based on deep learning according to claim 1, characterized in that: In step S02, the video image enhancement process includes illumination correction and contrast enhancement, wherein illumination correction adopts a histogram equalization algorithm, and contrast enhancement adopts an adaptive histogram equalization algorithm, and the specific formula is: ,in, is the input image, and are the mean and standard deviation of the image, respectively. It is the contrast intensity factor, and its value range is 1.5-3.

0.

4. The method for identifying driver fatigue driving status based on deep learning according to claim 1, characterized in that: In step S03, the convolution kernel order of the spatiotemporal graph convolutional network (ST-GCN) is 3, which can effectively capture the spatiotemporal relationship in the video sequence, and the feature extraction formula is: ,in, is the input feature matrix, is the adjacency matrix, is the degree matrix, is the learnable weight parameter, is the activation function is the order of the convolution kernel; the quantum deep learning adopts a quantum convolutional neural network and utilizes the superposition and entanglement characteristics of quantum bits to significantly improve the efficiency and accuracy of feature extraction. The feature extraction formula is: in, is the input quantum state, is the output quantum state, The adversarial network performs quantum convolution operation on the environmental perception data to generate realistic virtual environment data. and the discriminator Both use a deep convolutional neural network structure, where the output resolution of the generator is no less than 1920×1080 pixels to ensure the high quality and fidelity of the generated data; the loss functions of the generator and discriminator are defined as follows: The loss function of the generator for: ,in, From the prior distribution The noise vector sampled in , is the generator according to the noise vector The data generated, is the discriminator; the loss function of the discriminator for: ,in, From the real data distribution The data sampled in is the result of the discriminator’s judgment on the real data. is the discriminator’s judgment result on the generated data, is the discriminator’s judgment result on the generated data.

5. The method for identifying driver fatigue driving status based on deep learning according to claim 1, characterized in that: In step S04, the multimodal heterogeneous data fusion framework adopts a heterogeneous graph neural network, and the node feature update adopts a graph attention network (GAT) architecture. The specific formula is: ,in, For Node In the The characteristics of the layer, For Node The set of neighbor nodes of is the attention weight, is the learnable weight matrix, is the activation function.

6. The method for identifying driver fatigue driving status based on deep learning according to claim 1, characterized in that: In step S05, the deep learning model based on the Transformer architecture adopts an encoder-decoder structure, the number of encoder layers is 6, the number of decoder layers is 6, and the number of hidden units in each layer is 512; The meta-learning mechanism adopts model-independent meta-learning and can quickly adapt to new tasks and environments.

7. The method for identifying driver fatigue driving status based on deep learning according to claim 1, characterized in that: In step S06, the fatigue state judgment threshold is dynamically adjusted according to the driver's historical data and real-time data, and the threshold adjustment step is 0.01; the alarm signal includes a visual alarm, an auditory alarm and a tactile alarm, the visual alarm uses a red flashing light, the auditory alarm uses a 85-90dB buzzer, and the tactile alarm uses seat vibration.

8. The method for identifying driver fatigue driving status based on deep learning according to claim 1, characterized in that: In step S07, the reinforcement learning mechanism adopts a deep Q network, the learning rate α is 0.001, and the discount factor γ is 0.99; the online learning mechanism adopts an incremental learning algorithm, and the sample size for each update is 10-50; the quantum reinforcement learning adopts a quantum Q-learning algorithm, and the number of quantum bits is 5-10.

9. The method for identifying driver fatigue driving status based on deep learning according to claim 1, characterized in that: In step S08, the fatigue driving state recognition result is output through the vehicle control system, and the vehicle control system takes corresponding safety measures according to the recognition result, including: when the recognition result is a fatigue state, the vehicle automatically decelerates to a safe speed, and the safe speed range is 20-60km / h; at the same time, the vehicle automatically turns on the automatic driving assistance function, including lane keeping assist and automatic braking assist.

Images