CN111898502A - Dangerous goods vehicle identification method, device, computer storage medium, and electronic equipment - Google Patents

Abstract

A method, device, computer storage medium, and electronic equipment for identifying dangerous goods vehicles, including: acquiring road images; identifying dangerous goods vehicles in the road images by using a multi-level series convolutional neural network obtained by pre-training; The network includes the first-level convolutional neural network, the second-level convolutional neural network and the third-level convolutional neural network. The first-level convolutional neural network identifies dangerous goods vehicles and their vehicle areas in road images; the second-level convolutional neural network The neural network recognizes the dangerous goods signs in the vehicle area according to the screenshots of the vehicle area; the third-level convolutional neural network identifies the vehicle type according to the screenshots of the vehicle area and judges the confidence level of the dangerous goods signs, which is obtained according to the vehicle type and the confidence level of the dangerous goods signs. Dangerous Goods Vehicle Probability. By adopting the solution in the present application, various types of dangerous goods signs on the body of dangerous goods can be accurately identified, and automatic, accurate and efficient intelligent identification of dangerous goods vehicles can be realized.

Description

Dangerous goods vehicle identification method, device, computer storage medium, and electronic equipment

technical field

The present application relates to intelligent transportation technology, and in particular, to a method and device for identifying vehicles with dangerous goods, a computer storage medium, and an electronic device.

Background technique

Dangerous goods transport vehicle is a special vehicle for transporting petrochemicals, explosives, firecrackers and other dangerous goods. Because of its high safety requirements and high accident hazards, its control is very strict.

Problems existing in the prior art:

The current management and control methods mainly rely on manpower, which is costly and inefficient. Large-scale and timely supervision cannot be achieved.

SUMMARY OF THE INVENTION

Embodiments of the present application provide a method and device for identifying dangerous goods vehicles, a computer storage medium, and an electronic device to solve the above-mentioned technical problems.

According to a first aspect of the embodiments of the present application, a method for identifying a dangerous goods vehicle is provided, including the following steps:

Get road images;

Identifying the dangerous goods vehicle in the road image by using the multi-stage series convolutional neural network obtained by pre-training;

Wherein, the multi-level series convolutional neural network includes a first-level convolutional neural network, a second-level convolutional neural network and a third-level convolutional neural network, and the first-level convolutional neural network identifies the road image Medium-dangerous goods vehicles and their vehicle areas; the second-level convolutional neural network identifies the dangerous goods signs in the vehicle area according to the screenshot of the vehicle area returned by the first-level convolutional neural network; the third-level volume The convolutional neural network identifies the vehicle type according to the screenshot of the vehicle area returned by the first-level convolutional neural network and judges the confidence level of the dangerous goods sign, and obtains the dangerous goods vehicle probability according to the vehicle type and the confidence level of the dangerous goods sign.

According to a second aspect of the embodiments of the present application, there is provided a vehicle identification device for dangerous goods, including:

an acquisition module for acquiring road images;

an identification module for identifying dangerous goods vehicles in the road image by using the multi-stage series convolutional neural network obtained by pre-training;

The two-stage convolutional neural network in series includes a first-stage convolutional neural network, a second-stage convolutional neural network and a third-stage convolutional neural network, and the first-stage convolutional neural network identifies the road image Medium-dangerous goods vehicles and their vehicle areas; the second-level convolutional neural network identifies the dangerous goods signs in the vehicle area according to the screenshot of the vehicle area returned by the first-level convolutional neural network; the third-level volume The convolutional neural network identifies the vehicle type according to the screenshot of the vehicle area returned by the first-level convolutional neural network and judges the confidence level of the dangerous goods sign, and obtains the dangerous goods vehicle probability according to the vehicle type and the confidence level of the dangerous goods sign.

According to a third aspect of the embodiments of the present application, a computer storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, implements the steps of the above method for identifying a vehicle with dangerous goods.

According to a fourth aspect of the embodiments of the present application, an electronic device is provided, including a memory and one or more processors, where the memory is used to store one or more programs; the one or more programs are When executed by the one or more processors, the above-mentioned method for identifying vehicles with dangerous goods is implemented.

Using the method, device, computer storage medium, and electronic equipment for identifying dangerous goods vehicles provided in the embodiments of the present application, based on convolutional neural network implementation, it is possible to accurately identify various types of dangerous goods signs on the body of dangerous goods, and realize fully automatic, accurate, Efficient and intelligent identification of dangerous goods vehicles.

Description of drawings

The drawings described herein are used to provide further understanding of the present application and constitute a part of the present application. The schematic embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation of the present application. In the attached image:

FIG. 1 shows a schematic flowchart of the implementation of the method for identifying dangerous goods vehicles in Embodiment 1 of the present application;

FIG. 2 shows a schematic diagram of a process of judging the probability of dangerous goods vehicles in Embodiment 1 of the present application;

FIG. 3 shows the vehicle type identification process in the first embodiment of the present application;

Fig. 4 shows the calculation process of the vehicle dangerous goods type score in the first embodiment of the present application;

FIG. 5 shows a schematic diagram of the self-learning process in Embodiment 1 of the present application;

FIG. 6 shows a schematic structural diagram of a vehicle identification device for dangerous goods in the second embodiment of the present application;

FIG. 7 shows a schematic structural diagram of an electronic device in Embodiment 4 of the present application.

Detailed ways

The solutions in the embodiments of the present application may be implemented in various computer languages, for example, the object-oriented programming language Java and the literal translation scripting language JavaScript, and the like.

In order to make the technical solutions and advantages of the embodiments of the present application more clear, the exemplary embodiments of the present application will be described in further detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and Not all embodiments are exhaustive. It should be noted that the embodiments in the present application and the features of the embodiments may be combined with each other in the case of no conflict.

Example 1

FIG. 1 shows a schematic flowchart of the implementation of the method for identifying a dangerous goods vehicle in Embodiment 1 of the present application.

As shown in the figure, the method for identifying vehicles with dangerous goods includes:

Step 101, obtaining a road image;

Step 102, using the multi-level series convolutional neural network obtained by pre-training to identify the dangerous goods vehicle in the road image;

Wherein, the multi-level series convolutional neural network includes a first-level convolutional neural network, a second-level convolutional neural network and a third-level convolutional neural network, and the first-level convolutional neural network identifies the road image Medium-dangerous goods vehicles and their vehicle areas; the second-level convolutional neural network identifies the dangerous goods signs in the vehicle area according to the screenshot of the vehicle area returned by the first-level convolutional neural network; the third-level volume The convolutional neural network identifies the vehicle type according to the screenshot of the vehicle area returned by the first-level convolutional neural network and judges the confidence level of the dangerous goods sign, and obtains the dangerous goods vehicle probability according to the vehicle type and the confidence level of the dangerous goods sign.

During specific implementation, the embodiment of the present application can identify and intercept the vehicle area in the road image, and then predict whether there is a relevant sign of dangerous goods in the intercepted vehicle area, so as to determine whether the vehicle is a dangerous goods vehicle.

FIG. 2 shows a schematic diagram of a process of judging the probability of a dangerous goods vehicle in Embodiment 1 of the present application.

As shown in the figure, the embodiment of the present application designs a series and combined network:

1) The first-level network identification camera returns the vehicle and position in the overall image;

2) Take a screenshot of the vehicle area for the returned vehicle position;

3) The second-level network predicts the location of suspected signs of dangerous goods on the screenshot of the vehicle;

4) Send the screenshots of vehicles and suspected signs of dangerous goods into the third-level network at the same time;

5) The third-level network is a combined network including several classification models. It classifies vehicle types (tank trucks, heavy-duty trucks, cars, non-motor vehicles, etc.) with a pair of vehicle screenshots, and obtains the probability of dangerous goods types for different vehicle types. ; Second, make confidence judgment on suspected danger signs;

6) Finally, the probability of dangerous goods vehicles is judged by comprehensive vehicle type and the confidence level of dangerous goods signs, for example: dangerous goods vehicle probability=0.6×dangerous goods sign probability+0.4×dangerous goods vehicle type probability.

Using the method for identifying dangerous goods vehicles provided in the embodiments of the present application, which is implemented based on a convolutional neural network, can accurately identify various types of dangerous goods signs on the body of dangerous goods, and realize fully automatic, accurate, and efficient intelligent identification of dangerous goods vehicles.

In one embodiment, the obtaining of road images includes:

Extract the surveillance video captured by the camera;

Each frame of road image is acquired according to the surveillance video.

In one embodiment, the dangerous goods signs include one or more of the following:

Dangerous brands, triangular dangerous goods signs, dangerous goods diamond signs, "burning"/"explosive"/"rotten" dangerous goods subdivision signs.

In one embodiment, the training process of the first-level convolutional neural network includes:

Label the location information of each vehicle in the training image to obtain training data including vehicle location information;

According to the training data including the vehicle position information, the first-level convolutional neural network is obtained by relying on gradient backpropagation for training; the input of the first-level convolutional neural network is a road image, and the output is the corresponding position of the vehicle in the road image information.

During specific implementation, the steps of training the first-level convolutional neural network may be as follows:

1. Annotation of training data. Label the location information (xmin, ymin, xmax, ymax) of each vehicle in the training image. In this way, a batch of training data for knowing the location information is obtained.

2. Convolutional neural network training. Using the labeled training data, the detection model of the convolutional neural network (not limited to the specific type) is trained by gradient backpropagation. The model has the ability to know the vehicle position (xmin, ymin, xmax, ymax) in the image.

3. Deploy the trained model. Provide the input picture of the road. After the model is executed, the corresponding position (xmin, ymin, xmax, ymax) of the vehicle in the input picture can be obtained.

In one embodiment, the training process of the second-level convolutional neural network includes:

Mark the location information and type information of each dangerous goods sign in the training vehicle image, and obtain the training data including the location information and type information of the dangerous goods sign;

According to the training data including the location information and type information of the dangerous goods signs, rely on gradient back propagation for training to obtain a second-level convolutional neural network; the input of the second-level convolutional neural network is a vehicle image, and the output is a vehicle The location information and type information of each dangerous goods symbol in the image.

During specific implementation, the steps of training the second-level convolutional neural network may be as follows:

1. Annotation of training data. Mark the location information (xmin, ymin, xmax, ymax) and type information (burning, explosion, rot, etc.) of the dangerous goods signs in the training vehicle picture. In this way, a batch of training data for knowing the location and type information is obtained.

2. Convolutional neural network training. Using the labeled training data, the detection model of the convolutional neural network (not limited to the specific type) is trained by gradient backpropagation. The model has the ability to know the position (xmin, ymin, xmax, ymax) and type (burning, explosion, rot, etc.) of the dangerous goods in the image.

3. Deploy the trained model. Provide vehicle image input. After the model is executed, the position (xmin, ymin, xmax, ymax) and type (burning, explosion, rot, etc.) of each dangerous goods sign in the input image can be obtained.

In one embodiment, the use of a pre-trained multi-stage series convolutional neural network to identify dangerous goods vehicles in the road image includes:

Identify the location information of the area where the vehicle is located in the road image by using the first-level convolutional neural network;

According to the position information of the area where the vehicle is located, intercept the vehicle area in the matrix of the road image as the vehicle image;

Inputting the vehicle image into the second-level convolutional neural network, and using the second-level convolutional neural network to identify the location information and type information of the dangerous goods signs;

Utilize the third-level convolutional neural network to identify the vehicle type according to the screenshot of the vehicle area returned by the first-level convolutional neural network and judge the confidence level of the dangerous goods sign;

The probability of dangerous goods vehicles is obtained according to the vehicle type and the confidence level of the dangerous goods signs.

In specific implementation, the working process of the multi-stage series convolutional neural network can be as follows:

1. The road pictures obtained by the monitoring facilities are passed through the first-level convolutional neural network to obtain vehicle location information.

2. Implemented by programming, according to the given picture and vehicle position information, the vehicle area in the picture matrix is cut out as a new picture, which is provided to the second-level convolutional neural network for use.

3. The second-level convolutional neural network inputs the screenshot of the vehicle picture, and infers the dangerous goods signs and type information;

4. The third-level convolutional neural network identifies the vehicle type according to the screenshot of the vehicle area returned by the first-level convolutional neural network and judges the confidence level of the dangerous goods sign;

5. Obtain the probability of dangerous goods vehicles according to the vehicle type and the confidence of the dangerous goods signs.

In one embodiment, the method further comprises:

When a dangerous goods vehicle is identified, a hazard warning is sent.

In specific implementation, when a dangerous goods vehicle is identified, the road or location of the dangerous goods vehicle can be located, and the road or location of the dangerous goods vehicle can be marked and tracked on the map to prompt the surroundings or pass through the road section other vehicles, pay attention to safety.

In one embodiment, the training process of the third-level convolutional neural network includes:

A vehicle type recognition model is obtained by training several training vehicle sample data;

Determine the confidence level of the hazmat markings for each type of vehicle based on all sampled hazmat marked vehicles.

FIG. 3 shows the vehicle type identification process in the first embodiment of the present application.

As shown in the figure, the embodiment of the present application constructs a vehicle type identification model, and the specific construction process may be as follows:

A batch of training vehicle data is classified into vehicle types by manual annotation, and its types can be divided into large tank trucks, heavy trucks, light trucks, cars, etc. Then, it is trained with a multi-class deep neural network to obtain a deep learning model that can intelligently identify which class the vehicle belongs to.

FIG. 4 shows the calculation process of the vehicle dangerous goods type score in the first embodiment of the present application.

As shown in the figure, the embodiment of the present application adopts a normalized probability calculation method, assuming that the total number of all sampled vehicles with signs of dangerous goods is N, and the four types of vehicles in the example (large tank truck, heavy goods vehicle, light truck, and car) The total number of vehicles with dangerous goods signs are n1, n2, n3, and n4 respectively. Then the scores of the four types of dangerous goods in vehicles are: n1/N, n2/N, n3/N, and n4/N. where N=n1+n2+n3+n4.

In one embodiment, the method further comprises:

When the dangerous goods vehicle is a first type of dangerous goods vehicle, obtain data of the dangerous goods vehicle, and retrain the multi-level convolutional neural network using the data of the dangerous goods vehicle as a training sample;

The dangerous goods vehicle of the first type is a vehicle type that accounts for less than the dangerous goods vehicle of the second type when training the multi-level convolutional neural network.

Considering that different vehicle types have different probabilities of belonging to dangerous goods vehicles, the embodiment of the present application adopts a real-time self-learning adjustment method to determine the probability that each type of vehicle belongs to dangerous goods vehicles;

The specific implementation method is: statistics the probability of detecting dangerous goods signs for various types of vehicles in the monitoring scene, update the ratio of vehicles with dangerous goods signs in various models in real time, and use this ratio to determine the probability of various types of vehicles belonging to dangerous goods vehicles. . The method provided by the embodiment of the present application has the self-learning ability for different types of dangerous goods vehicles in various regions, and has good generalization performance.

FIG. 5 shows a schematic diagram of the self-learning process in Embodiment 1 of the present application.

As shown in the figure, thanks to the addition of the vehicle type self-learning module, a positive feedback self-improvement effect is formed between the vehicle type self-learning branch and the dangerous goods sign recognition branch. For the hazmat sign recognition model, the performance of its hazmat sign recognition effect on various models is not the same, which is mainly affected by the proportional distribution of various models in the training data. For hazmat vehicles, the most It is a large tanker, which may account for more than 80%, and has sufficient training data, so its model performs better. But for light trucks, the proportion may be less than 5%, so the model training is biased towards vehicle types with more data, and the performance on light trucks may be very poor. After the introduction of the self-learning model, the vehicle type self-learning module and the dangerous goods sign identification module can be used in each vehicle database to obtain the data of light trucks and other types of dangerous goods vehicles in a targeted manner, so as to quickly supplement the training data with a small proportion. Vehicle type Dangerous goods vehicle, after balancing the data types, the dangerous goods sign recognition module can achieve better results on various vehicle types. In turn, the improvement of the dangerous goods sign identification module can improve the accuracy of the proportion of various types of vehicles in the vehicle type identification module. Positive feedback is formed to improve the overall frame effect.

To sum up, the embodiments of the present application have the following advantages:

1) Fully automatic, no manpower required. It is only necessary for cameras that have been set up at each intersection to capture road images, and the embodiment of the present application can automatically identify each vehicle in the image and whether the vehicle is a dangerous goods vehicle.

2) Accurate. Relying on the embodiment of the present application, the identification accuracy rate of dangerous goods vehicles is as high as nearly 100%, so that all dangerous goods vehicles passing through the camera on the road can be accurately identified and reported.

3) Efficient. The embodiments of the present application can be made into software packages and deployed on various servers to achieve an image processing speed of hundreds of thousands of frames per hour, and to meet monitoring needs in a wide range.

4) Timely. In the embodiment of the present application, the dangerous goods vehicle that appears can be identified in time at the first time when the front-end camera returns an image. To discover hidden dangers in a timely manner, prevent problems before they occur, and avoid problems that are only discovered after an accident.

Embodiment 2

Based on the same inventive concept, an embodiment of the present application provides a vehicle identification device for dangerous goods. The principle of the device for solving technical problems is similar to that of a method for identifying vehicles for dangerous goods, and repeated details will not be repeated.

FIG. 6 shows a schematic structural diagram of a vehicle identification device for dangerous goods in Embodiment 2 of the present application.

As shown in the figure, the dangerous goods vehicle identification device includes:

an acquisition module 601, configured to acquire road images;

An identification module 602, configured to identify dangerous goods vehicles in the road image by using the multi-stage series convolutional neural network obtained by pre-training;

Wherein, the multi-level series convolutional neural network includes a first-level convolutional neural network, a second-level convolutional neural network and a third-level convolutional neural network, and the first-level convolutional neural network identifies the road image Medium-dangerous goods vehicles and their vehicle areas; the second-level convolutional neural network identifies the dangerous goods signs in the vehicle area according to the screenshot of the vehicle area returned by the first-level convolutional neural network; the third-level volume The convolutional neural network identifies the vehicle type according to the screenshot of the vehicle area returned by the first-level convolutional neural network and judges the confidence level of the dangerous goods sign, and obtains the dangerous goods vehicle probability according to the vehicle type and the confidence level of the dangerous goods sign.

Using the dangerous goods vehicle identification device provided in the embodiment of the present application, implemented based on a convolutional neural network, can accurately identify various types of dangerous goods signs on the dangerous goods body, and realize fully automatic, accurate and efficient intelligent identification of dangerous goods vehicles.

In one embodiment, the acquisition module includes:

A video extraction unit for extracting surveillance video captured by a camera;

An image acquisition unit, configured to acquire each frame of road image according to the surveillance video.

In one embodiment, the dangerous goods signs include one or more of the following:

Dangerous brands, triangular dangerous goods signs, dangerous goods diamond signs, "burning"/"explosive"/"rotten" dangerous goods subdivision signs.

In one embodiment, the training process of the first-level convolutional neural network includes:

Label the location information of each vehicle in the training image to obtain training data including vehicle location information;

According to the training data including the vehicle position information, the first-level convolutional neural network is obtained by relying on gradient backpropagation for training; the input of the first-level convolutional neural network is a road image, and the output is the corresponding position of the vehicle in the road image information.

In one embodiment, the training process of the second-level convolutional neural network includes:

Mark the location information and type information of each dangerous goods sign in the training vehicle image, and obtain the training data including the location information and type information of the dangerous goods sign;

According to the training data including the location information and type information of the dangerous goods signs, rely on gradient back propagation for training to obtain a second-level convolutional neural network; the input of the second-level convolutional neural network is a vehicle image, and the output is a vehicle The location information and type information of each dangerous goods symbol in the image.

In one embodiment, the use of a pre-trained two-stage series convolutional neural network to identify dangerous goods vehicles in the road image includes:

Identify the location information of the area where the vehicle is located in the road image by using the first-level convolutional neural network;

According to the position information of the area where the vehicle is located, intercept the vehicle area in the matrix of the road image as the vehicle image;

Inputting the vehicle image into the second-level convolutional neural network, and using the second-level convolutional neural network to identify the location information and type information of the dangerous goods signs;

Identify the vehicle type according to the screenshot of the vehicle area returned by the first-level convolutional neural network and determine the confidence level of the dangerous goods sign;

The probability of dangerous goods vehicles is obtained according to the vehicle type and the confidence level of the dangerous goods signs.

In one embodiment, the training process of the third-level convolutional neural network includes:

A vehicle type recognition model is obtained by training several training vehicle sample data;

Determine the confidence level of the hazmat markings for each type of vehicle based on all sampled hazmat marked vehicles.

In one embodiment, the apparatus further comprises:

The early warning module is used to send a danger warning when a dangerous vehicle is identified.

In one embodiment, the apparatus further comprises:

The self-learning module is configured to acquire the data of the dangerous goods vehicle when the dangerous goods vehicle is the first type of dangerous goods vehicle, and use the data of the dangerous goods vehicle as a training sample to retrain the multi-level convolution Neural Networks;

The dangerous goods vehicle of the first type is a vehicle type that accounts for less than the dangerous goods vehicle of the second type when training the multi-level convolutional neural network.

Embodiment 3

Based on the same inventive concept, an embodiment of the present application further provides a computer storage medium, which will be described below.

The computer storage medium stores a computer program thereon, and when the computer program is executed by the processor, implements the steps of the method for identifying a dangerous goods vehicle according to the first embodiment.

Using the computer storage medium provided in the embodiments of the present application, based on the convolutional neural network implementation, it is possible to accurately identify various types of dangerous goods signs on the dangerous goods body, and realize fully automatic, accurate and efficient intelligent identification of dangerous goods vehicles.

Embodiment 4

Based on the same inventive concept, an embodiment of the present application further provides an electronic device, which will be described below.

FIG. 7 shows a schematic structural diagram of an electronic device in Embodiment 4 of the present application.

As shown, the electronic device includes a memory 701, and one or more processors 702, the memory is used to store one or more programs; the one or more programs are stored by the one or more processors When executed, the method for identifying vehicles with dangerous goods as described in the first embodiment is implemented.

Using the electronic equipment provided in the embodiments of the present application, implemented based on a convolutional neural network, can accurately identify various types of dangerous goods signs on the dangerous goods body, and realize fully automatic, accurate and efficient intelligent identification of dangerous goods vehicles.

As will be appreciated by those skilled in the art, the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

While the preferred embodiments of the present application have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be construed to include the preferred embodiment and all changes and modifications that fall within the scope of this application.

Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims (10)

1. a dangerous goods vehicle identification method, is characterized in that, comprises: Get road images; Identifying the dangerous goods vehicle in the road image by using the multi-stage series convolutional neural network obtained by pre-training; Wherein, the multi-stage series convolutional neural network includes a first-stage convolutional neural network, a second-stage convolutional neural network and a third-stage convolutional neural network, and the first-stage convolutional neural network identifies the road image Medium-dangerous goods vehicles and their vehicle areas; the second-level convolutional neural network identifies the dangerous goods signs in the vehicle area according to the screenshot of the vehicle area returned by the first-level convolutional neural network; the third-level volume The convolutional neural network identifies the vehicle type according to the screenshot of the vehicle area returned by the first-level convolutional neural network and judges the confidence level of the dangerous goods sign, and obtains the dangerous goods vehicle probability according to the vehicle type and the confidence level of the dangerous goods sign. 2. The method according to claim 1, wherein the acquiring a road image comprises: Extract the surveillance video captured by the camera; Each frame of road image is acquired according to the surveillance video. 3. The method according to claim 1, wherein the dangerous goods signs comprise one or more of the following: Dangerous brands, triangular dangerous goods signs, dangerous goods diamond signs, "burning"/"explosive"/"rotten" dangerous goods subdivision signs. 4. The method according to claim 1, wherein the training process of the first-level convolutional neural network comprises: Label the location information of each vehicle in the training image to obtain training data including vehicle location information; According to the training data including the vehicle position information, the first-level convolutional neural network is obtained by relying on gradient backpropagation for training; the input of the first-level convolutional neural network is a road image, and the output is the corresponding position of the vehicle in the road image information. 5. The method according to claim 1, wherein the training process of the second-level convolutional neural network comprises: Mark the location information and type information of each dangerous goods sign in the training vehicle image, and obtain the training data including the location information and type information of the dangerous goods sign; According to the training data including the location information and type information of the dangerous goods signs, rely on gradient back propagation for training to obtain a second-level convolutional neural network; the input of the second-level convolutional neural network is a vehicle image, and the output is a vehicle The location information and type information of each dangerous goods symbol in the image. 6. The method according to claim 1, wherein the training process of the third-level convolutional neural network comprises: A vehicle type recognition model is obtained by training several training vehicle sample data; Determine the confidence level of the hazmat markings for each type of vehicle based on all sampled hazmat marked vehicles. 7. The method of claim 1, further comprising: When the dangerous goods vehicle is a first type of dangerous goods vehicle, obtain data of the dangerous goods vehicle, and retrain the multi-level convolutional neural network using the data of the dangerous goods vehicle as a training sample; The dangerous goods vehicle of the first type is a vehicle type that accounts for less than the dangerous goods vehicle of the second type when training the multi-level convolutional neural network. 8. A vehicle identification device for dangerous goods, comprising: an acquisition module for acquiring road images; an identification module for identifying dangerous goods vehicles in the road image by using the multi-stage series convolutional neural network obtained by pre-training; Wherein, the multi-level series convolutional neural network includes a first-level convolutional neural network, a second-level convolutional neural network and a third-level convolutional neural network, and the first-level convolutional neural network identifies the road image the vehicle and its vehicle area; the second-level convolutional neural network identifies the presence of dangerous goods signs in the vehicle area according to the screenshot of the vehicle area returned by the first-level convolutional neural network; the third-level convolutional neural network The network identifies the vehicle type according to the screenshot of the vehicle area returned by the first-level convolutional neural network and judges the confidence level of the dangerous goods sign, and obtains the dangerous goods vehicle probability according to the vehicle type and the confidence level of the dangerous goods sign. 9 . A computer storage medium, characterized in that a computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented. 10. An electronic device, comprising a memory and one or more processors, wherein the memory is used to store one or more programs; the one or more programs are controlled by the one or more processors When executed, the method as claimed in any one of claims 1 to 7 is implemented.

Images