A vehicle management method and system for an intelligent construction site

Abstract

This application relates to image processing technology, and specifically to a vehicle management method and system for a smart construction site. The method includes the steps of: obtaining a dataset containing vehicle washing images; using environmental images of different weather conditions as style images; calculating a content weighting coefficient and a style weighting coefficient based on the vehicle washing images and the style images; setting a pre-trained convolutional neural network, inputting the vehicle washing images and the style images, adding the product of the style weighting coefficient and the style loss to the product of the content weighting coefficient and the content loss as a loss function, iteratively processing the vehicle washing images by minimizing the loss function, obtaining a style transfer image after the iteration is completed, and expanding the dataset for training a vehicle cleanliness recognition model based on the style transfer image. This application has the effect of improving the expanded data and quality.

Description

Technical Field

[0001] This application relates to image processing technology, and specifically to a vehicle management method and system for intelligent construction sites. Background Art

[0002] To prevent construction vehicles from carrying sediment in the construction site onto urban roads, it is necessary to supervise and manage the construction vehicles entering and leaving the construction site to ensure that the construction vehicles are effectively washed before leaving the construction site and reduce the impact of construction on air quality. To reduce labor costs, when supervising and managing vehicles, a vehicle non-washing capture system is usually used to supervise and manage whether the vehicle has been washed.

[0003] To improve the efficiency of vehicle supervision and management, it is usually judged whether the vehicle is clean during the vehicle washing process, and a neural network is used to identify and judge whether the vehicle is clean by recognizing images. For example, the patent application document with the publication number CN116758521A discloses a vehicle cleanliness detection method and device. The method includes the steps of: obtaining an image of the vehicle to be detected; inputting the image of the vehicle to be detected into the batch linear mapping module in the pre-constructed vehicle cleanliness detection model, dividing the image of the vehicle to be detected into multiple image blocks, and the vehicle cleanliness detection model is obtained by training the model with sample images labeled with cleanliness labels, and inputting the multiple image blocks in sequence; determining whether the vehicle cleanliness is qualified according to the output result of the vehicle cleanliness detection model.

[0004] Due to limited training samples, there may be a situation of insufficient recognition accuracy, so it is necessary to expand the data set. The style transfer algorithm can apply the style of one image to another image while retaining the content features of the target image. The style transfer algorithm can be used to generate images under different weather conditions to expand the data set.

[0005] However, the style transfer algorithm has the problem of how to balance content loss and style loss. Assigning unreasonable weights to content loss and style loss may affect the quality of the generated vehicle washing images in multiple environments and affect the recognition of vehicle washing effects. Summary of the Invention

[0006] To solve the problem of how to balance content loss and style loss, this application provides a vehicle management method and system for intelligent construction sites.

[0007] In a first aspect, this application provides a vehicle management method for intelligent construction sites, adopting the following technical solution:

[0008] A vehicle management method for an intelligent construction site, comprising the steps of: obtaining a dataset containing vehicle cleaning images, and calculating the content complexity and style feature vector of the vehicle cleaning images; using environmental images of different weather conditions as style images, and calculating the content complexity and style feature vector of the style images; calculating a content weighting coefficient; the content weighting coefficient is in a direct proportional relationship with the content complexity of the vehicle cleaning images and in an inverse proportional relationship with the content complexity of the style images; calculating a style weighting coefficient; wherein, obtaining the similarity degree between the style image vector and the style feature vector, the style weighting coefficient is in an inverse proportional relationship with the similarity degree and in an inverse proportional relationship with the content complexity of the vehicle cleaning images; setting a pre-trained convolutional neural network, inputting the vehicle cleaning images and the style images, adding the product of the style weighting coefficient and the style loss to the product of the content weighting coefficient and the content loss as a loss function, iterating the vehicle cleaning images by minimizing the loss function, obtaining a style transfer image after the iteration is completed, and augmenting the dataset for training the vehicle cleanliness recognition model according to the style transfer image.

[0009] The beneficial effects are as follows: The content complexity indicates the amount of information contained in the image, and the feature vector quantifies the similarity degree between images. Calculate the style loss weight and content loss weight according to the content complexity and style feature vector of the image, and dynamically adjust the weight ratio of the content loss and style loss for each pair of images used in style transfer according to the characteristics of the vehicle cleaning images and style images in the original dataset, so as to ensure that the image obtained by style transfer can not only retain the details and structure in the original image, but also accurately integrate the illumination scene and weather scene of the target into the generated image, and finally generate a natural, delicate and style transfer image that meets the specific scene requirements. Augmenting the dataset according to the style transfer image is beneficial to improving the subsequent vehicle cleaning recognition accuracy and the accuracy of vehicle cleaning management.

[0010] Optionally, the calculation formula for the content complexity of the vehicle cleaning images is: , where represents the content complexity of the th vehicle cleaning image, represents the maximum value of the pixel point gradient value in the th vehicle cleaning image, represents the average value of the pixel point gradient value in the th vehicle cleaning image, represents the number of pixel points with a gradient value greater than in the th vehicle cleaning image; represents the number of pixel points in the th vehicle cleaning image, is the The number of types of gray values that appear in a vehicle washing image; similar to the calculation formula for the content complexity of a vehicle washing image, the content complexity of a style image is obtained.

[0011] The beneficial effects are as follows: It is the difference between the overall gradient magnitude of the image and the maximum gradient value; It is the relative magnitude of the number of pixel points with prominent gradients in the image; It is the relative magnitude of the number of types of gray values that appear in the image; the content complexity is comprehensively quantified through the product of different-dimensional characteristic factors.

[0012] Optionally, the content complexity of a vehicle washing image is: the reciprocal of the variance of the bar heights of the gray histogram of the vehicle washing image.

[0013] The beneficial effects are as follows: The height of each bar in the gray histogram represents the number of pixels corresponding to the gray value, and the variance reflects the degree of deviation of these gray values from their mean. The larger the value, the more dispersed the pixel gray values in the image, that is, the richer the texture of the image, the higher the contrast, and the more obvious the image details. The smaller the value, the more concentrated the pixel gray values in the image, that is, the smoother the texture of the image, the lower the contrast, and the less obvious the image details.

[0014] Optionally, the mathematical expression of the style feature vector is: , is the style feature vector of the th style image, is the th number of pixel points corresponding to the , is the total number of pixel point eigenvalue obtained by local binary pattern. The number of pixel point eigenvalues obtained by local binary pattern with different parameters is different; similar to the construction method of the style feature vector of the style image, the style feature vector of the vehicle washing image is constructed.

[0015] The beneficial effects are as follows: Since local binary pattern can extract the texture features in the image well, the texture features in images with different style features are also differently reflected. The differences in style between such images can be reflected in the eigenvalue distributions of all pixel points in the image. In some style images, the eigenvalue distributions are relatively average, and in some style images, the eigenvalue distributions are relatively concentrated. Therefore can reflect the style features of images with different styles.

[0016] Optionally, the calculation formula for the style weighting coefficient is: , represents the th vehicle washing image and the The style weighting coefficient used for style transfer of the denotes the exponential function with as the base, represents the style feature vector of the vehicle washing images in the dataset, represents the style feature vector of the style image, denotes the th content complexity of the vehicle washing image, represents the cosine similarity function, denotes and the degree of similarity.

[0017] The beneficial effect is: is the degree of style similarity between the vehicle washing image and the style image. Normalizing gives . The larger the value of , the greater the degree of style similarity between the two images, indicating that the weather scenarios or lighting conditions corresponding to the two images are more similar. At this time, there is no need to make excessive adjustments to the generation of the image in terms of style features. Instead, the style loss weight should be reduced to retain more content information in the original vehicle washing image in the new image.

[0018] Optionally, the calculation formula for the style weighting coefficient is: . denotes the style weighting coefficient used for style transfer of the th vehicle washing image and the th style image, represents the normalization function, represents the Euclidean distance, denotes the exponential function with as the base, represents the style feature vector of the vehicle washing images in the dataset, represents the style feature vector of the style image, denotes the th content complexity of the vehicle washing image, denotes and the degree of similarity.

[0019] The beneficial effect is: represents the normalization function, represents the Euclidean distance, denotes and the degree of similarity. The purpose of normalization is to scale the data proportionally. By calculating the Euclidean distance between vectors, the degree of style similarity between images can be quantified.

[0020] Optionally, the calculation formula for the content weighting coefficient is: , where represents the content weighting coefficient used when performing style transfer between the th vehicle cleaning image and the th style image, represents the exponential function with as the base, represents the content complexity of the th vehicle cleaning image, represents the content complexity of the th style image.

[0021] In a second aspect, the present application provides a vehicle management system for a smart construction site, adopting the following technical solution:

[0022] A vehicle management system for a smart construction site includes: a processor and a memory, where the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the vehicle management method for a smart construction site described above is implemented.

[0023] The beneficial effect is: generating a computer program for the vehicle management method for a smart construction site described above and storing it in the memory to be loaded and executed by the processor. Thus, making a system according to the memory and the processor is convenient to use.

[0024] The present application has the following technical effects:

[0025] 1. The content complexity indicates the amount of information contained in the image, and the feature vector quantifies the similarity degree between images. Calculating the style loss weight and the content loss weight according to the content complexity of the image and the style feature vector, and dynamically adjusting the weight ratio of the content loss and the style loss for each pair of images used for style transfer according to the characteristics of the vehicle cleaning images and the style images in the original dataset, so as to ensure that the image obtained by style transfer can not only retain the details and structure in the original image, but also accurately integrate the illumination scene and the weather scene of the target into the generated image, and finally generate a style transfer image that is natural, delicate and meets the specific scene requirements. Expanding the dataset according to the style transfer image is beneficial to the effect of improving the subsequent vehicle cleaning recognition accuracy and is beneficial to improving the accuracy of vehicle cleaning management.

[0026] 2. The large complexity of the content of vehicle cleaning images means that vehicle cleaning images contain more details. The higher the content complexity of the images during vehicle cleaning, the more precisely these complex details need to be retained to avoid losing the structural information of the engineering vehicle cleaning images during the style transfer process. Therefore, the content loss weight used during the style transfer of vehicle cleaning images should be larger. For vehicle cleaning images with smaller content complexity, since the simple content is already relatively clear and not easily distorted, the weight of content loss can be appropriately reduced so that more weight can be placed on style loss to achieve the effect of more lighting conditions and weather scene transfers;

[0027] Since style transfer emphasizes the style of the style image rather than the content, when the content complexity of the style image is greater, it means that the image contains more details, structures, and complex visual elements. In this process, if the content complexity of the style image is high, the details and structures of the original vehicle cleaning image are likely to be affected by the details of the style image. If the weight of content loss is too large at this time, it will limit the effective transfer of the style features of the style image. Therefore, it is necessary to reduce the weight of content loss so that the generated image can better integrate into the complex structure and details of the style image;

[0028] When the content complexity of the style image is small, the style image itself is relatively simple and may present relatively blurred and single image features (such as foggy days or dark days). At this time, if the weight of content loss is too low, the generated image may lack the basic structure of the source image, resulting in the lack of recognition of the vehicle cleaning image in the generated image. Therefore, in order to ensure that the content such as the structure and object form of the source image is not erased too much and the generated image retains the essence of the source image, it is necessary to increase the weight of content loss so that the generated image better retains the content features of the source image, thereby achieving a more natural and coherent style transfer effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] By referring to the detailed description below with reference to the drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present application will become readily understandable. In the drawings, several embodiments of the present application are shown in an exemplary rather than restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts.

[0030] Figure 1 is a flowchart of a method for vehicle management in a smart construction site according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0032] It should be understood that when terms such as "first" and "second" are used in the claims, the description, and the drawings of the present application, they are only used to distinguish different objects and are not used to describe a specific order. The terms "including" and "comprising" used in the description and claims of the present application indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0033] An embodiment of the present application discloses a vehicle management method for a smart construction site. Refer to Figure 1 , including the steps:

[0034] S1: Obtain a dataset containing vehicle washing images, and calculate the content complexity and style feature vectors of the vehicle washing images; use environmental images of different weather conditions as style images, and calculate the content complexity and style feature vectors of the style images.

[0035] Obtain an existing dataset of vehicle washing images, and collect style images under different lighting conditions and different weather conditions for generating vehicle images under different weather and different lighting scenarios. Exemplarily, both the vehicle washing images and the style images are grayscale images. Use the Sobel operator to calculate the gradient of each pixel point in the image, and count the number of types of gray values that appear in the image. Exemplarily, if only pixel points with a gray value of 0 appear in the image, then this value is 1; if pixel points with gray values of 0, 3, 4, and 5 appear, then this value is 4.

[0036] Since different images have different detailed information, when using the style transfer algorithm for image style transfer, different degrees of image details in the vehicle washing images will be retained in the generated new images. To avoid the loss or blurring of the image details in the vehicle washing images in the new images, the present application calculates the content complexity of the images according to the pixel points in the dataset images.

[0037] In one embodiment, the calculation formula for the content complexity of the vehicle washing images is:

[0038] , where represents the content complexity of the th vehicle washing image, represents the The maximum value of the gradient value of the pixel points in the vehicle washing image, which represents the mean value of the gradient value of the pixel points in the vehicle washing image of the th vehicle washing image; which represents the number of pixel points in the vehicle washing image of the th vehicle washing image, where the gradient value is greater than

[0039] ; represents the difference between the overall gradient magnitude and the maximum gradient value of the

[0040] th

[0041] vehicle washing image. The larger this value is, the smaller the overall gradient of the image is, the less detailed information is in the image, and the smaller the content complexity of the image is. The smaller this value is, the larger the overall gradient of the image is, the more detailed information is in the image, and the larger the content complexity of the image is.

[0042] represents the relative magnitude of the number of pixel points with prominent gradients in the image. The larger this value is, the more pixel points with prominent gradients are in the image, the more detailed information is in the image, and the larger the content complexity of the image is. The smaller this value is, the fewer pixel points with prominent gradients are in the image, the less detailed information is in the image, and the smaller the content complexity of the image is.

[0043] represents the relative magnitude of the number of types of gray values that appear in the image. The larger this value is, the more colors appear in the image, the more detailed information is in the image, and the larger the content complexity of the image is. The smaller this value is, the fewer colors appear in the image, the less detailed information is in the image, and the smaller the content complexity of the image is.

[0044] Similar to the calculation formula of the content complexity of vehicle cleaning images, the content complexity of the style image is obtained. Details are not elaborated here.

[0045] In one embodiment, taking the style feature vector of the style image as an example, the mathematical expression of the style feature vector is: , is the style feature vector of the th style image, is the th number of pixel points corresponding to the th eigenvalue in the , is the total number of pixel point eigenvalue obtained by local binary pattern. The number of pixel point eigenvalues obtained by local binary pattern with different parameters is different.

[0046] For example, if using local binary pattern with a radius of 1 and based on 8 sampling points, then , if using local binary pattern with a radius of 1 and based on 4 sampling points, then .

[0047] Since local binary pattern can extract texture features in images well, and the texture features in images with different style features are also differently reflected. The difference in style between such images can be reflected in the eigenvalue distribution of all pixel points in the image. In some style images, the eigenvalue distribution is relatively average, while in some style images, the eigenvalue distribution is relatively concentrated. Therefore can reflect the style features of images with different styles.

[0048] Similar to the construction method of the style feature vector of the style image, the style feature vector of the vehicle cleaning image is constructed. Details are not elaborated here.

[0049] S2: Calculate the content weighting coefficient; the content weighting coefficient is in a direct proportional relationship with the content complexity of the vehicle cleaning image and in an inverse proportional relationship with the content complexity of the style image.

[0050] The calculation formula of the content weighting coefficient is: , where represents the content weighting coefficient used when performing style transfer between the th vehicle cleaning image and the th style image, represents the exponential function with as the base, represents the content complexity of the th vehicle cleaning image, represents the content complexity of the th style image.

[0051] S3: Calculate the style weighting coefficient; wherein, obtain the similarity degree between the style image vector and the style feature vector, and the style weighting coefficient is inversely proportional to the similarity degree and inversely proportional to the content complexity of the vehicle cleaning image.

[0052] The calculation formula for the style weighting coefficient is: , denotes the th vehicle cleaning image and the th style image when performing style transfer, and the style weighting coefficient used is denotes the exponential function with as the base, denotes the style feature vector of the vehicle cleaning images in the dataset, denotes the style feature vector of the style image, denotes the th content complexity of the vehicle cleaning image, is the cosine similarity function, denotes and similarity degree.

[0053] A large content complexity of the vehicle cleaning image means that the vehicle cleaning image contains more details. Then, the higher the content complexity of the image during vehicle cleaning, the more precisely these complex details need to be retained to avoid losing the structural information of the engineering vehicle cleaning image during the style transfer process. Therefore, the content loss weight used for the th vehicle cleaning image style transfer should be larger. For vehicle cleaning images with smaller content complexity, since the simple content is already relatively clear and not easily distorted, the weight of the content loss can be appropriately reduced so that more weight can be placed on the style loss to achieve the effect of more lighting conditions and weather scene transfers.

[0054] Since style transfer emphasizes the style of the style image rather than the content, when the content complexity of the style image is larger, it means that the image contains more details, structures, and complex visual elements. In this process, if the content complexity of the style image is high, the details and structures of the original vehicle cleaning image are easily affected by the details of the style image. If the weight of the content loss is too large at this time, it will limit the effective transfer of the style features of the style image. Therefore, it is necessary to reduce the weight of the content loss so that the generated image can better integrate into the complex structure and details of the style image.

[0055] When the content complexity of the style image is small, the style image itself is relatively simple and may exhibit relatively blurred and single image features (such as foggy days or dark days). At this time, if the content loss weight is too low, the generated image may lack the basic structure of the source image, resulting in a lack of recognition of the vehicle washing image. Therefore, in order to ensure that the content such as the structure and object shape of the source image is not overly erased and the generated image retains the essence of the source image, it is necessary to increase the weight of the content loss so that the generated image better retains the content features of the source image, thereby achieving a more natural and coherent style transfer effect.

[0056] is the style similarity degree between the vehicle washing image and the style image. Since the value range of the cosine similarity is from -1 to 1, for the convenience of calculation, in the formula, is normalized by adding 1 and dividing by 2.

[0057] The larger is, the greater the style similarity degree between the two images indicates, and the closer the weather scenes or lighting conditions corresponding to the two images are. At this time, there is no need to make excessive adjustments to the generation of the image in terms of style features. At this time, the style loss weight should be reduced so that the content information in the original vehicle washing image can be more retained in the new image. The smaller this value is, the greater the style gap between the two images indicates, and the greater the difference in the weather scenes or lighting conditions corresponding to the two images. At this time, more adjustments need to be made to the generation of the image in terms of style features. At this time, the style loss weight should be increased so that the style information in the target style image can be more reflected in the new image.

[0058] At the same time, the greater the content complexity of the vehicle washing image, the more details the image has. If too much style loss is imposed during style transfer, it may cause detail loss or unnatural transitions in the new image. Therefore, the weight of the style loss should be reduced to avoid excessive changes in the image style. When the content complexity of the vehicle washing image is smaller, it means that the image has less detail information and it is not easy to lose too much content during the style transfer process. Therefore, a larger style loss weight can be used to make the generated image better absorb the style information in the lighting conditions or weather scenes of the target style image.

[0059] In other embodiments, the calculation formula of the style weighting coefficient can also be:

[0060] , represents the style weighting coefficient used when performing style transfer on the th vehicle washing image and the th style image. denotes the exponential function with as the base, denotes the style feature vector of the vehicle washing images in the dataset, denotes the style feature vector of the style image, denotes the content complexity of the -th vehicle washing image.

[0061] denotes the normalization function, denotes the Euclidean distance, denotes the similarity degree between and . The purpose of normalization is to scale the data proportionally so that it falls within a small specific interval, such as [0, 1]. is similar to the above effect and will not be elaborated here.

[0062] S4: Set up a pre-trained convolutional neural network, input the vehicle washing image and the style image, sum the product of the style weighting coefficient and the style loss and the product of the content weighting coefficient and the content loss as the loss function, iterate the vehicle washing image by minimizing the loss function, and after the iteration is completed, obtain the style transfer image, and expand the dataset for training the vehicle cleanliness recognition model according to the style transfer image.

[0063] Using the style transfer algorithm (VGG19 network, Inception network or other feature extraction networks can be used) to generate images is the prior art and will not be elaborated here. In this process, the style loss and the content loss are continuously calculated, and by multiplying the style loss weight by the style loss and multiplying the content loss weight by the content loss, and then performing subsequent training iterations, the generated image can not only well retain the content of the vehicle washing process in the original image, but also make the image fuse different illuminations and different weather conditions. By performing image style transfer through this method, a higher-quality and richer vehicle washing recognition dataset can be generated. By training the vehicle washing recognition neural network with the expanded dataset, the accuracy of vehicle washing recognition can be improved. The training and use of the vehicle washing recognition neural network are the prior art and will not be elaborated here.

[0064] The embodiment of the present application also discloses a vehicle management system for a smart construction site, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the vehicle management method for a smart construction site according to the present application is implemented.

[0065] The above system also includes other components well known to those skilled in the art such as a communication bus and a communication interface. Their settings and functions are known in the art, so they will not be elaborated here.

[0066] In this application, the aforementioned memory can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic storage medium or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high bandwidth memory (HBM), hybrid memory cube (HMC), and so on, or any other medium that can be used to store the required information and can be accessed by an application program, module, or both. Any such computer storage medium can be part of the device or accessible or connectable to the device.

[0067] Although this specification has shown and described multiple embodiments of this application, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will think of many changes, alterations, and alternative ways without departing from the spirit and scope of this application. It should be understood that various alternatives to the embodiments of this application described herein can be employed in the practice of this application.

[0068] The above are all the preferred embodiments of this application, and the protection scope of this application is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of this application shall be covered within the protection scope of this application.

Claims

1. A vehicle management method for a smart construction site, characterized in that: Includes steps: Obtain a data set containing vehicle washing images, calculate the content complexity and style feature vector of the vehicle washing images; use environmental images of different weather conditions as style images, and calculate the content complexity and style feature vector of the style images; Calculate the content weighting coefficient; the content weighting coefficient is directly proportional to the content complexity of the vehicle washing image and inversely proportional to the content complexity of the style image; Calculating a style weighting coefficient; wherein the similarity between the style feature vector of the vehicle washing image and the style feature vector of the style image is obtained, and the style weighting coefficient is inversely proportional to the similarity and inversely proportional to the content complexity of the vehicle washing image; A pre-trained convolutional neural network is set up, and the vehicle cleaning image and style image are input. The product of the style weighting coefficient and the style loss is added to the product of the content weighting coefficient and the content loss as the loss function. The vehicle cleaning image is iterated by minimizing the loss function. After the iteration, the style transfer image is obtained. According to the style transfer image, the data set used to train the vehicle cleanliness recognition model is expanded.

2. The vehicle management method for a smart construction site according to claim 1, characterized in that: The calculation formula for the content complexity of the vehicle washing image is: ,in, Indicates The content complexity of the vehicle washing image, Indicates The maximum value of the pixel gradient value in the vehicle washing image. Indicates The mean value of the pixel gradient in the vehicle washing image, Indicates In the vehicle cleaning image, the gradient value is greater than The number of pixels; Indicates The number of pixels in the vehicle washing image, For the The number of grayscale value types that appear in a vehicle washing image; The content complexity of the style image is obtained in the same way as the calculation formula of the content complexity of the vehicle washing image.

3. The vehicle management method for a smart construction site according to claim 1, characterized in that: The content complexity of the vehicle washing image is: the inverse of the variance of the grayscale histogram column height of the vehicle washing image.

4. The vehicle management method for a smart construction site according to any one of claims 1 to 3, characterized in that: The mathematical expression of the style feature vector is: , For the The style feature vector of the style image, For the The style image The number of pixels corresponding to the eigenvalues, , is the total number of pixel feature values ​​obtained by the local binary pattern. The number of pixel feature values ​​obtained by using local binary patterns with different parameters is different. The style feature vector of the vehicle washing image is constructed in the same way as the method for constructing the style feature vector of the style image.

5. The vehicle management method for a smart construction site according to claim 4, characterized in that: The calculation formula of style weighting coefficient is: , Indicates The vehicle cleaning image The style weighting coefficient used when performing style transfer on the style image. Indicates The exponential function with base , represents the style feature vector of the vehicle washing image in the dataset, The style feature vector representing the style image, Indicates The content complexity of the vehicle washing image, represents the cosine similarity function, express and degree of similarity.

6. The vehicle management method for a smart construction site according to claim 1, characterized in that: The calculation formula of content weight coefficient is: ,in, Indicates The vehicle cleaning image The content weighting coefficient used when performing style transfer on the style image. Indicates The exponential function with base , Indicates The content complexity of the vehicle washing image, Indicates The content complexity of the style image.

7. A vehicle management system for a smart construction site, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the vehicle management method for a smart construction site according to any one of claims 1-6 is implemented.

Images