Image acquisition device for vehicle identification based on deep learning and identification method

Abstract

The invention discloses a vehicle recognition method, and particularly relates to an image acquisition device for vehicle recognition based on deep learning and a recognition method.The image acquisition device comprises a first connecting rod, a camera is arranged at the lower end of the first connecting rod, a first rain baffle is arranged on the first connecting rod, and a sun shield is arranged below the first rain baffle; the sun shield is connected to the first connecting rod through a first hinge. The device is not only provided with the camera for image acquisition, but also provided with the sun shield and the rain shield, so that the influence of severe weather such as direct sunlight, rainwater and the like on the shooting quality of the camera is effectively prevented, and the stability and definition of image acquisition are improved. The two-way convolutional neural network model is adopted for license plate number recognition, the license plate information can be accurately and rapidly recognized through the deep learning technology, and the recognition accuracy and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of new generation information technology, and in particular to an image acquisition device and a recognition method for vehicle recognition based on deep learning. Background Art

[0002] With the acceleration of urbanization and the continuous increase in traffic flow, vehicle recognition technology is increasingly widely used in the fields of traffic management, security monitoring, and intelligent parking. Traditional vehicle recognition methods mainly rely on manual observation or simple image processing technology. These methods have problems such as low recognition efficiency, low accuracy, and susceptibility to environmental interference, which are difficult to meet the needs of modern traffic management. In the process of vehicle recognition, image acquisition is a crucial link. However, adverse factors such as light changes, rain, and haze in the natural environment often lead to a decrease in the quality of the collected images, which in turn affects the subsequent recognition effect. In addition, there are differences in license plate formats, colors, fonts, etc. in different regions, which also brings challenges to vehicle recognition. In order to overcome these problems, deep learning technology has been gradually applied to the field of vehicle recognition in recent years. Deep learning can automatically learn feature information in images by constructing complex neural network models to achieve accurate recognition of vehicles and license plates. However, the performance of deep learning models depends largely on the quality of the input images. Therefore, how to collect high-quality vehicle images in complex environments has become the key to the development of vehicle recognition technology. In this context, the present invention proposes an image acquisition device and recognition method for vehicle recognition based on deep learning. Summary of the invention

[0003] In order to solve the above-mentioned problems in the prior art, the present invention provides an image acquisition device and recognition method for vehicle recognition based on deep learning to solve the current technical problems.

[0004] The technical solution adopted by the present invention to solve its technical problem is: The present invention provides an image acquisition device and recognition method for vehicle identification based on deep learning, comprising: a first connecting rod, a camera is arranged at the lower end of the first connecting rod, a first rain shield is arranged on the first connecting rod, a sun visor is arranged below the first rain shield, and the sun visor is connected to the first connecting rod through a first hinge.

[0005] Preferably, the sun visor includes two parts, a first telescopic rod is connected between the sun visor and the first connecting rod, one end of the first telescopic rod is connected to the sun visor through a second hinge, and the other end of the first telescopic rod is connected to the first connecting rod through a third hinge, and the axes of the first hinge, the second hinge and the third hinge are parallel to each other.

[0006] Preferably, there is a gap between the two sun visors, and the first rain shield is located directly above the gap.

[0007] Preferably, the first connecting rod is connected to the main shaft of the motor at the top, the housing of the motor is connected to the second connecting rod, the second connecting rod is provided with a second rain shield, and the second rain shield is located directly above the motor.

[0008] Preferably, the second connecting rod is connected to the third connecting rod, the third connecting rod is connected to the second telescopic rod, and the second telescopic rod is connected to the base.

[0009] Preferably, a first reinforcing rib is fixed below the third connecting rod, and the first reinforcing rib is also connected to the second telescopic rod and the second connecting rod.

[0010] Preferably, a second reinforcing rib is connected above the third connecting rod, and the second reinforcing rib is connected to the second telescopic rod.

[0011] Preferably, the identification method includes the following: Step 1: Data collection: Use a camera to collect photos of passing vehicles. The vehicle photos must include the license plate. Step 2: Image recognition: Get the vehicle photo collected by the camera and identify the vehicle's license plate number; Step 3: Output: Output the recognized license plate number in text form.

[0012] Preferably, in the Step 2: image recognition step, a dual-path convolutional neural network model is used to recognize the vehicle's license plate number.

[0013] The beneficial effects of the present invention are: Multifunctional design: The device is not only equipped with a camera for image acquisition, but also designed with a sun visor and rain shield, which effectively prevents the influence of bad weather, such as direct sunlight, rain, etc., on the camera shooting quality, and improves the stability and clarity of image acquisition.

[0014] Flexible adjustable sun visor: The sun visor is connected to the first connecting rod through the first telescopic rod and multiple hinges, and the angle and position can be flexibly adjusted to adapt to different light conditions, further enhancing the adaptability of image acquisition.

[0015] Stable structure: The design of the first and second reinforcing ribs enhances the overall stability of the device, which can effectively reduce vibration and shaking, and ensure the stability of image acquisition, especially in harsh environments or high-speed movements.

[0016] Efficient image recognition method: A dual-path convolutional neural network model is used for license plate number recognition. This deep learning technology can accurately and quickly identify license plate information, improving recognition accuracy and efficiency.

[0017] High integration: The entire device has a compact structure and integrates functions such as image acquisition, sunshade and rain shelter, flexible adjustment and efficient recognition. It is easy to install and maintain and is suitable for vehicle recognition needs in various scenarios.

[0018] Easy to expand and upgrade: Due to the use of deep learning technology, this recognition method is easy to train models and adjust parameters according to actual needs to adapt to license plate formats and styles in different regions, and it is also convenient for future technology upgrades and expansions.

[0019] High degree of automation: From data collection to image recognition and then to result output, the entire process is highly automated, reducing manual intervention and improving work efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which: Figure 1 It is a structural schematic diagram of an image acquisition device and a recognition method for vehicle recognition based on deep learning according to an embodiment of the present invention; Figure 2 It is a structural schematic diagram of an image acquisition device and a recognition method for vehicle recognition based on deep learning according to an embodiment of the present invention; Figure 3 It is a structural schematic diagram of an image acquisition device and a recognition method for vehicle recognition based on deep learning according to an embodiment of the present invention; Figure 4 It is a structural schematic diagram of an image acquisition device and a recognition method for vehicle recognition based on deep learning according to an embodiment of the present invention; Figure 5 It is a structural schematic diagram of an image acquisition device and a recognition method for vehicle recognition based on deep learning according to an embodiment of the present invention; Figure 6 It is a schematic diagram of an image acquisition device and a recognition method for vehicle recognition based on deep learning according to an embodiment of the present invention.

[0021] Description of reference numerals: exist Figure 1-Figure 6 In the figure, the camera 1; the second telescopic rod 2; the sun visor 3; the first reinforcing rib 4; the third connecting rod 5; the second reinforcing rib 6; the second connecting rod 7; the second rain shield 8; the motor 9; the first connecting rod 10; the first rain shield 11; and the first telescopic rod 12. 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] Embodiment 1: The present invention provides an image acquisition device and a recognition method for vehicle recognition based on deep learning, comprising: a first connecting rod 10, a camera 1 is arranged at the lower end of the first connecting rod 10, a first rain shield 11 is arranged on the first connecting rod 10, a sun visor 3 is arranged below the first rain shield 11, and the sun visor 3 is connected to the first connecting rod 10 through a first hinge. The first connecting rod 10, the second connecting rod 7 and the third connecting rod 5 are all metal rods, the first rain shield 11 and the second rain shield 8 can use stainless steel metal plates or plastic plates, the sun visor 3 can also use plastic plates, and it is recommended to use metal plates.

[0024] The sun visor 3 includes two pieces, and a first telescopic rod 12 is connected between the sun visor 3 and the first connecting rod. One end of the first telescopic rod 12 is connected to the sun visor 3 through a second hinge, and the other end of the first telescopic rod 12 is connected to the first connecting rod 10 through a third hinge. The axes of the first hinge, the second hinge and the third hinge are parallel to each other. The first telescopic rod 12 is an electric telescopic rod with a power-off self-locking function.

[0025] There is a gap between the two sunshades 3, and the first rain shield 11 is located right above the gap.

[0026] The first connecting rod 10 is connected to the main shaft of the motor 9 at the top, the housing of the motor 9 is connected to the second connecting rod 7, and the second connecting rod 7 is provided with a second rain shield 8, which is located just above the motor 9. The motor 9 can use a steering gear.

[0027] The second connecting rod 7 is connected to the third connecting rod 5, the third connecting rod 5 is connected to the second telescopic rod 2, and the second telescopic rod 2 is connected to the base. The second telescopic rod 2 can also be replaced by a metal rod, and the base is a metal plate, which can be fixed to the ground with anchor bolts.

[0028] A first reinforcing rib 4 is fixed below the third connecting rod 5, and the first reinforcing rib 4 is also connected to the second telescopic rod 2 and the second connecting rod 7. A second reinforcing rib 6 is connected above the third connecting rod 5, and the second reinforcing rib 6 is connected to the second telescopic rod 2. Both the first reinforcing rib 4 and the second reinforcing rib 6 are metal plates.

[0029] This application includes the following identification methods: Step 1: Data collection: Use camera 1 to collect photos of passing vehicles. The vehicle photos must include the license plate. Step 2: Image recognition: Get the vehicle photo collected by camera 1 and identify the vehicle license plate number; Step 3: Output: Output the recognized license plate number in text form.

[0030] Step 2: In the image recognition step, a dual-channel convolutional neural network model is used to recognize the vehicle’s license plate number.

[0031] The size of the captured image is 1024x1024, and this image is used as the input of the model.

[0032] The specific structure of the dual-path convolutional neural network model includes: Branch 1: Convolution module 11: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolution module 12: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 13: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 14: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 15: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 16: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; The feature map output by branch 1 has a size of 16x16 and 64 feature maps, which are marked as T1, T2, T3, ..., T63, T64.

[0033] Branch 2: Convolutional module 21: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 5x5, the stride is 1, and the padding is 2; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 22: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 5x5, the stride is 1, and the padding is 2; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 23: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 5x5, the stride is 1, and the padding is 2; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 24: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 5x5, the stride is 1, and the padding is 2; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 25: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 5x5, the stride is 1, and the padding is 2; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 26: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 5x5, the stride is 1, and the padding is 2; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; The feature map output by branch 2 has a size of 16x16 and 64 feature maps, marked as P1, P2, P3, ..., P63, P64.

[0034] The feature fusion module includes: 64 fused feature maps, marked as R1, R2, R3, ... R63, R64, each feature map has a size of 16x16, where: The generation method of the 32 feature maps R1-R32 is the same. Each feature map is generated by first randomly selecting one from the 64 feature maps T1-T64, then randomly selecting one from the 64 feature maps P1-P64, and finally adding the corresponding position elements of the two feature maps to obtain the fused feature map; The generation method of the 16 feature maps R33-R48 is the same. Each feature map is generated by randomly selecting two from the 64 feature maps T1-T64, adding the corresponding position elements of the two feature maps, and obtaining the fused feature map; The generation method of the 16 feature maps R49-R64 is the same. The generation method of each feature map is to randomly select two from the 64 feature maps P1-P64, add the elements at corresponding positions of the two feature maps, and obtain the fused feature map.

[0035] Multilayer Perceptron: First, the 64 features of size 16x16 are flattened into a vector of 16384 dimensions; Linear layer 1: The number of input features is 16384, the number of output features is 32768, and there is no bias; Linear layer 2: The number of input features is 32768, the number of output features is N, and there is no bias term, where N represents the number of license plate categories used for training and testing; As a preferred solution, the Dropout technology is used to add a Dropout layer between the fully connected layers to reduce overfitting.

[0036] The last layer is the classification layer, which uses the Softmax activation function to output the classification probability.

[0037] Learning rate: set to 0.0001, which is determined through tuning to better balance the model convergence speed and stability.

[0038] Batch size: 128 samples are processed per iteration to adapt to the limitations of video memory and computing resources.

[0039] Optimizer: The Adam optimizer is used, which can automatically adjust the learning rate and has better effects on sparse gradients and non-stationary targets.

[0040] Loss function: The cross entropy loss function is used, which is suitable for multi-classification tasks.

[0041] Number of training rounds: set to 100 rounds. Experiments have verified that this setting can enable the model to fully converge.

[0042] The ratio of training set to test set is 9:1, and 90% of the data are randomly selected for training.

[0043] For example, we used 10,000 photos of 100 vehicles taken at different times during training and testing, and the test accuracy was 99%.

[0044] Embodiment 2: The present invention provides an image acquisition device and a recognition method for vehicle recognition based on deep learning, comprising: a first connecting rod 10, a camera 1 is arranged at the lower end of the first connecting rod 10, a first rain shield 11 is arranged on the first connecting rod 10, a sun visor 3 is arranged below the first rain shield 11, and the sun visor 3 is connected to the first connecting rod 10 through a first hinge. The first connecting rod 10, the second connecting rod 7 and the third connecting rod 5 are all metal rods, the first rain shield 11 and the second rain shield 8 can use stainless steel metal plates or plastic plates, the sun visor 3 can also use plastic plates, and it is recommended to use metal plates.

[0045] The sun visor 3 includes two pieces, and a first telescopic rod 12 is connected between the sun visor 3 and the first connecting rod. One end of the first telescopic rod 12 is connected to the sun visor 3 through a second hinge, and the other end of the first telescopic rod 12 is connected to the first connecting rod 10 through a third hinge. The axes of the first hinge, the second hinge and the third hinge are parallel to each other. The first telescopic rod 12 is an electric telescopic rod with a power-off self-locking function.

[0046] There is a gap between the two sunshades 3, and the first rain shield 11 is located right above the gap.

[0047] The first connecting rod 10 is connected to the main shaft of the motor 9 at the top, the housing of the motor 9 is connected to the second connecting rod 7, and the second connecting rod 7 is provided with a second rain shield 8, which is located just above the motor 9. The motor 9 can use a steering gear.

[0048] The second connecting rod 7 is connected to the third connecting rod 5, the third connecting rod 5 is connected to the second telescopic rod 2, and the second telescopic rod 2 is connected to the base. The second telescopic rod 2 can also be replaced by a metal rod, and the base is a metal plate, which can be fixed to the ground with anchor bolts.

[0049] A first reinforcing rib 4 is fixed below the third connecting rod 5, and the first reinforcing rib 4 is also connected to the second telescopic rod 2 and the second connecting rod 7. A second reinforcing rib 6 is connected above the third connecting rod 5, and the second reinforcing rib 6 is connected to the second telescopic rod 2. Both the first reinforcing rib 4 and the second reinforcing rib 6 are metal plates.

[0050] This application includes the following identification methods: Step 1: Data collection: Use camera 1 to collect photos of passing vehicles. The vehicle photos must include the license plate. Step 2: Image recognition: Get the vehicle photo collected by camera 1 and identify the vehicle license plate number; Step 3: Output: Output the recognized license plate number in text form.

[0051] Step 2: In the image recognition step, a dual-channel convolutional neural network model is used to recognize the vehicle’s license plate number.

[0052] The size of the captured image is 1024x1024, and this image is used as the input of the model.

[0053] The specific structure of the dual-path convolutional neural network model includes: Branch 1: Convolution module 11: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 12: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 13: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 14: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 15: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 16: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; The feature map output by branch 1 has a size of 16x16 and 64 feature maps, which are marked as T1, T2, T3, ..., T63, T64.

[0054] Branch 2: Convolutional module 21: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 7x7, the stride is 1, and the padding is 3; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 22: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 7x7, the stride is 1, and the padding is 3; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 23: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 7x7, the stride is 1, and the padding is 3; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 24: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 5x5, the stride is 1, and the padding is 2; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 25: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 5x5, the stride is 1, and the padding is 2; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; Convolutional module 26: Convolutional layer: The number of convolution kernels is 2, the convolution kernel size is 3x3, the stride is 1, and the padding is 1; Activation function: RELU; Pooling layer: type is maximum pooling, pooling size is 2x2, and stride is 2; The feature map output by branch 2 has a size of 16x16 and 64 feature maps, marked as P1, P2, P3, ..., P63, P64.

[0055] The feature fusion module includes: 64 fused feature maps, marked as R1, R2, R3, ... R63, R64, each feature map has a size of 16x16, where: The generation method of the 32 feature maps R1-R32 is the same. Each feature map is generated by first randomly selecting one from the 64 feature maps T1-T64, then randomly selecting one from the 64 feature maps P1-P64, and finally adding the corresponding position elements of the two feature maps to obtain the fused feature map; The generation method of the 16 feature maps R33-R48 is the same. Each feature map is generated by randomly selecting two from the 64 feature maps T1-T64, adding the corresponding position elements of the two feature maps, and obtaining the fused feature map; The generation method of the 16 feature maps R49-R64 is the same. The generation method of each feature map is to randomly select two from the 64 feature maps P1-P64, add the elements at corresponding positions of the two feature maps, and obtain the fused feature map.

[0056] Multilayer Perceptron: First, the 64 features of size 16x16 are flattened into a vector of 16384 dimensions; Linear layer 1: The number of input features is 16384, the number of output features is 32768, and there is no bias; Linear layer 2: The number of input features is 32768, the number of output features is N, and there is no bias term, where N represents the number of license plate categories used for training and testing; As a preferred solution, the Dropout technology is used to add a Dropout layer between the fully connected layers to reduce overfitting.

[0057] The last layer is the classification layer, which uses the Softmax activation function to output the classification probability.

[0058] Learning rate: set to 0.0001, which is determined through tuning to better balance the model convergence speed and stability.

[0059] Batch size: 256 samples are processed per iteration to adapt to the limitations of video memory and computing resources.

[0060] Optimizer: The Adam optimizer is used, which can automatically adjust the learning rate and has better effects on sparse gradients and non-stationary targets.

[0061] Loss function: The cross entropy loss function is used, which is suitable for multi-classification tasks.

[0062] Number of training rounds: set to 100 rounds. Experiments have verified that this setting can enable the model to fully converge.

[0063] The ratio of training set to test set is 9:1, and 90% of the data are randomly selected for training.

[0064] For example, we used 10,000 photos of 100 vehicles taken at different times during training and testing, and the test accuracy was 99.2%.

[0065] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An image acquisition device and recognition method for vehicle recognition based on deep learning, comprising: The first connecting rod is characterized in that: a camera is arranged at the lower end of the first connecting rod, a first rain shield is arranged on the first connecting rod, a sun visor is arranged below the first rain shield, and the sun visor is connected to the first connecting rod through a first hinge.

2. The image acquisition device and recognition method for vehicle recognition based on deep learning according to claim 1, characterized in that: The sun visor includes two parts, and a first telescopic rod is connected between the sun visor and the first connecting rod. One end of the first telescopic rod is connected to the sun visor through a second hinge, and the other end of the first telescopic rod is connected to the first connecting rod through a third hinge. The axes of the first hinge, the second hinge and the third hinge are parallel to each other.

3. The image acquisition device and recognition method for vehicle recognition based on deep learning according to claim 2, characterized in that: There is a gap between the two sun visors, and the first rain shield is located directly above the gap.

4. The image acquisition device and recognition method for vehicle recognition based on deep learning according to claim 1, characterized in that: The upper part of the first connecting rod is connected to the main shaft of the motor, the housing of the motor is connected to the second connecting rod, the second connecting rod is provided with a second rain shield, and the second rain shield is located directly above the motor.

5. The image acquisition device and recognition method for vehicle recognition based on deep learning according to claim 4, characterized in that: The second connecting rod is connected to the third connecting rod, the third connecting rod is connected to the second telescopic rod, and the second telescopic rod is connected to the base.

6. The image acquisition device and recognition method for vehicle recognition based on deep learning according to claim 5, characterized in that: A first reinforcing rib is fixed below the third connecting rod, and the first reinforcing rib is also connected to the second telescopic rod and the second connecting rod.

7. The image acquisition device and recognition method for vehicle recognition based on deep learning according to claim 5, characterized in that: A second reinforcing rib is connected above the third connecting rod, and the second reinforcing rib is connected to the second telescopic rod.

8. The image acquisition device and recognition method for vehicle recognition based on deep learning according to claim 1, characterized in that: The identification methods include: Step 1: Data collection: Use a camera to collect photos of passing vehicles. The vehicle photos must include the license plate. Step 2: Image recognition: Get the vehicle photo collected by the camera and identify the vehicle's license plate number; Step 3: Output: Output the recognized license plate number in text form.

9. The image acquisition device and recognition method for vehicle recognition based on deep learning according to claim 8, characterized in that: In the Step 2: image recognition step, a dual-path convolutional neural network model is used to recognize the vehicle license plate number.

Images