An image recognition-based road surface state intelligent sensing method and system
By integrating the ESCA attention mechanism and the LCAhead module into the EfficientNetB0 model, the ESCA-LCANet model is constructed, which solves the problems of insufficient road feature recognition capability and poor robustness in the existing technology, and realizes real-time and accurate road condition recognition and safe driving guidance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANTONG UNIV
- Filing Date
- 2026-01-20
- Publication Date
- 2026-06-05
Smart Images
Figure CN122157187A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent transportation road surface perception technology, and more specifically, relates to a method and system for intelligent perception of road surface conditions based on image recognition. Background Technology
[0002] With the rapid development of intelligent transportation systems and autonomous driving technologies, road environment perception has become a crucial link in ensuring traffic safety and stable vehicle operation. Road surface conditions, as an important reflection of the state of transportation infrastructure, not only directly affect vehicle driving safety and ride comfort but also relate to road maintenance decisions and lifecycle management. Therefore, constructing an image classification model capable of identifying road surface types and defect characteristics in real time is of significant research value and application significance for achieving intelligent identification and automated management of road conditions.
[0003] Existing technologies often have the following problems: (1) Insufficient ability to identify multi-dimensional fine-grained road features (such as dryness, wetness, snow accumulation, water accumulation, unevenness, etc.); (2) Redundant model parameters and large computational load, making it difficult to achieve real-time inference on embedded devices or mobile terminals; (3) Poor robustness in complex environments such as multiple lighting conditions, rain, snow, and shadows, resulting in inaccurate classification results; (4) Lack of a precise and quantitative risk classification and corresponding driving guidelines, making it impossible to provide drivers with clear action instructions in a scientific and accurate manner.
[0004] Therefore, it is very important to invent a method and system for intelligent perception of road conditions based on image recognition. Summary of the Invention
[0005] This invention provides a method and system for intelligent perception of road conditions based on image recognition, in order to solve the above-mentioned problems.
[0006] According to one aspect of the present invention, a road surface condition intelligent sensing system based on image recognition is provided, comprising: The image preprocessing module is used to capture images and perform preprocessing operations, divide the training set, validation set and test set, and perform data augmentation operations; A road condition intelligent perception module that integrates the ESCA attention mechanism and LCAhead is based on the EfficientNetB0 model and integrates the ESCA attention mechanism and LCAhead module. Model training module; Image classification and recognition module; And a risk classification and driving guide module.
[0007] Furthermore, in the road condition intelligent perception module that integrates the ESCA attention mechanism and LCAhead, the EfficientNetB0 network model consists of convolutional layers, multiple MBConv modules, average pooling layers, max pooling layers, and softmax fully connected layers. The SE attention mechanism in the original MBConv module is replaced with the ESCA attention mechanism, and the last 1×1 convolutional module is replaced with the LCAhead module. While reducing parameters, lightweight spatial mixing, spatial self-attention mechanism, and residual reconstruction module are introduced to improve the ability to distinguish subtle road features.
[0008] Furthermore, the model training module configures training hyperparameters, uses the cross-entropy loss function, trains the ESCA-LCANet road environment perception model using the training dataset, calculates and updates the weights of the ESCA-LCANet road environment perception model in each iteration, and at the end of each training round, uses a validation set to evaluate the performance of the ESCA-LCANet road environment perception model, observing the accuracy and loss value of the validation set so that training stops when the model converges, preventing overfitting.
[0009] Furthermore, the image classification and recognition module uses a test set to evaluate the classification and recognition performance of the ESCA-LCANet road environment perception model, calculates the model's precision, accuracy, and F1 score on the test set, and constructs a visualization of the confusion matrix based on the F1 score, number of parameters, computational cost, and precision. This aims to systematically evaluate the model's classification performance and error distribution in real-world scenarios, and to detect the model's lightweight nature and required computational power.
[0010] Furthermore, the risk classification and driving guide module constructs a precise and quantifiable risk classification and corresponding driving guide, transforming vague experience into clear standards.
[0011] According to another aspect of the present invention, a road surface condition intelligent perception method based on image recognition is provided, comprising the following steps: cleaning the publicly available RSCD dataset and selecting road environment images with typical features to obtain a road environment perception dataset of 21 categories to be used. Data augmentation was performed on the data images. The acquired RSCD road environment status dataset was divided proportionally according to an 8:1:1 ratio to obtain a training dataset, a validation dataset, and a test dataset. Based on the EfficientNetB0 model, an improved ESCA-LCANet road environment perception model is constructed by integrating the ESCA attention mechanism and the LCAhead module. Configure the training hyperparameters and update the model weights based on the road environment perception dataset to obtain the trained ESCA-LCANet road environment perception model; The data is input into the ESCA-LCANet road environment perception model for training. The performance of the road environment perception dataset is evaluated and classified to identify different road environment states. The weights are then deployed to mobile devices to achieve real-time classification and recognition of road environment states.
[0012] Furthermore, data augmentation is performed on the data images, including random rotation, random scaling, random flipping, and brightness adjustment.
[0013] Furthermore, based on the EfficientNetB0 model, and by integrating the ESCA attention mechanism and the LCAhead module, the specific steps for constructing the improved ESCA-LCANet road environment perception model are as follows: ESCA module: An enhancement module that combines efficient channel attention and spatial pyramid structure is embedded in the backbone network to enhance the model's ability to extract and focus on multi-scale road features; LCAhead module: Design a lightweight classification head that reduces the number of parameters and computational complexity while maintaining accuracy; Transfer learning is performed on pre-trained weights loaded onto large datasets to accelerate convergence; Configure training hyperparameters and update model weights on the training set; During training, performance is monitored using a validation set to prevent overfitting.
[0014] The specific steps for constructing the improved ESCA-LCANet road environment perception model based on the EfficientNetB0 model, integrating the ESCA attention mechanism and the LCAhead module, are as follows: ESCA module: An enhancement module that combines efficient channel attention and spatial pyramid structure is embedded in the backbone network to enhance the model's ability to extract and focus on multi-scale road features; LCAhead module: Design a lightweight classification head that reduces the number of parameters and computational complexity while maintaining accuracy; Transfer learning is performed on pre-trained weights loaded onto large datasets to accelerate convergence; Configure training hyperparameters and update model weights on the training set; During training, performance is monitored using a validation set to prevent overfitting.
[0015] Risk Classification and Driving Guide Module: Constructs precise and quantifiable risk classifications and corresponding driving guidelines, transforming vague experiences into clear standards.
[0016] According to another aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of an image recognition-based intelligent road condition sensing method of the present invention.
[0017] According to another aspect of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of an image recognition-based intelligent road condition sensing method of the present invention.
[0018] Compared with existing technologies, the beneficial effects of the above-described method of the present invention are as follows: This invention can improve the ability to identify multi-dimensional fine-grained road surface features, and achieve accurate differentiation and comprehensive judgment of various types of road surface features such as dry and wet conditions, snow accumulation, water accumulation, and unevenness. This invention can reduce model complexity and computational cost. Through structural optimization and lightweight design, it reduces parameter redundancy and computational load, enabling the model to achieve efficient, real-time inference on embedded devices or mobile terminals. This invention can enhance the robustness and generalization ability of the model in complex environments, improve its stability and recognition accuracy in variable scenarios such as multiple lighting conditions, rain, snow, and shadows, and ensure the reliability of road surface condition classification results.
[0019] This invention can construct precise and quantifiable risk classifications and corresponding driving guidelines, transforming vague experiences into clear standards. It provides drivers with clear action instructions, and in conjunction with intelligent perception systems, enables a shift from passive response to proactive prevention, significantly improving driving safety. Attached Figure Description
[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments will be briefly described below. Obviously, the drawings described below only relate to some embodiments of the present invention and are not intended to limit the present invention.
[0021] Figure 1 This is a schematic diagram of the method flow in a preferred embodiment of the present invention; Figure 2 This is a system flowchart in a preferred embodiment of the present invention; Figure 3 This is a schematic diagram of the ESCA-LCANet model structure in a preferred embodiment of the present invention; Figure 4 This is a schematic diagram of the ESCA attention module in a preferred embodiment of the present invention. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention.
[0023] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.
[0024] Example 1: This invention provides an intelligent road condition sensing system based on image recognition, comprising: The image preprocessing module is used to capture images and perform preprocessing operations, divide the image into training, validation and test sets, perform data augmentation operations, reduce interference caused by external factors, and simulate road environment conditions more realistically. A road condition intelligent perception module that integrates the ESCA attention mechanism and LCAhead; Model training module; Image classification and recognition module; Risk classification and driving guide module.
[0025] The specific implementation process is as follows: Data cleaning was performed on the publicly available RSCD dataset, and road environment images with typical characteristics were selected to obtain a road environment perception dataset of 21 categories to be used. Data augmentation was performed on the data images, including random rotation, random scaling, random flipping, and brightness adjustment. The acquired RSCD road environment status dataset was divided proportionally according to an 8:1:1 ratio to obtain training dataset, validation dataset, and test dataset. Based on the EfficientNetB0 model, an improved ESCA-LCANet road environment perception model is constructed by integrating the ESCA attention mechanism and the LCAhead module. Configure the training hyperparameters and update the model weights based on the road environment perception dataset to obtain the trained ESCA-LCANet road environment perception model; The data is input into the ESCA-LCANet road environment perception model for training. The performance of the road environment perception dataset is evaluated and classified to identify different road environment states. The weights are then deployed to mobile devices to achieve real-time classification and recognition of road environment states. like Figure 1-2 As shown, the adhesion effect of driving on this type of road is obtained based on the road environment conditions, and detailed driving suggestions are obtained to improve driving safety.
[0026] 1. Image Preprocessing Module: First, the road environment perception dataset is divided into training, validation, and test sets in an 8:1:1 ratio. The training set contains 12,801 images, the validation set contains 1,601 images, and the test set contains 1,598 images. Specifically: the training set contains labels and target variable values for algorithm training and building the prediction model; the validation set is used to adjust model hyperparameters, evaluate model performance, and select the best model and hyperparameters after training; the test set is used to predict model performance and provide accurate evaluation results. After image preprocessing, the images are uniformly cropped to 224 pixels × 224 pixels × 3 channels.
[0027] Data augmentation is performed on the data images, including rotation, cropping, translation, and color adjustment, to achieve diverse transformations of the training data.
[0028] 2. Road surface condition intelligent sensing module: Based on the EfficientNetB0 model, this project integrates the ESCA attention mechanism and the LCAhead module. Specifically, the EfficientNetB0 network model consists of convolutional layers, multiple MBConv modules, average pooling layers, max pooling layers, and softmax fully connected layers. The key to this convolutional neural network, which integrates the ESCA attention mechanism and the LCAhead module, is replacing the SE attention mechanism in the original MBConv module with the ESCA attention mechanism, and replacing the last 1×1 convolutional module with the LCAhead module. This reduces the number of parameters while introducing lightweight spatial mixing, spatial self-attention mechanisms, and a residual reconstruction module to improve the ability to distinguish subtle road features. The specific model structure diagram is shown below. Figure 3 As shown.
[0029] Specifically, the ESCA attention mechanism consists of parallel channel subspace attention and spatial aggregation attention, combined with a cross-modal fusion mechanism. This enhances the model's ability to perceive fine-grained spatial features of the road environment while maintaining low computational cost, thereby significantly improving the model's recognition accuracy and generalization performance for different road types and environmental changes. Figure 4 As shown.
[0030] The core idea of channel subspace attention is to divide the original feature channels into several subspaces and independently model the dependencies between channels within each subspace, thereby achieving finer-grained feature selection.
[0031] Let the input features be The specific calculation process of the channel subspace attention module is as follows: (1) Global feature description: GAP and GMP represent global average pooling and global max pooling, respectively, used to perceive different road environment characteristics.
[0032] (2) Subspace partitioning and convolution modeling: Divide Y into S subspaces according to the channel dimension. One-dimensional convolution is used within each subspace to capture local channel correlations:
[0033] in, This represents the Sigmoid activation function.
[0034] (3) Concatenate all subspace attention weights to obtain the global channel weight map:
[0035] The core of the spatial convergence attention module lies in obtaining the spatial correlation and geometric structure features of the road scene, thereby improving the model's discrimination ability and spatial perception in complex environments.
[0036] Let the input features be The specific calculation process of the SPA module is as follows: (1) Directional pooling: In order to capture the spatial statistical features in both the horizontal and vertical directions at the same time, global average pooling is performed in both dimensions: ,
[0037] (2) Feature aggregation and convolution modeling: The pooling results from the two directions are concatenated in the spatial dimension:
[0038] (3) Depthwise separable convolution is used to extract local spatial context information, then normalization and ReLU activation function are used to regularize the data distribution, and finally... Convolution achieves channel fusion:
[0039] in, This represents the Sigmoid activation function, used to generate spatial attention map representations. .
[0040] The core of the LCAhead module lies in using a lightweight cross-channel feature aggregation module to improve the ability to express local texture and global spatial dependencies, thereby enhancing the ability to identify road environment conditions.
[0041] Let the input features be The specific calculation process of the LCAHead module is as follows: (1) Perform depthwise convolution on each channel while keeping the number of channels constant: , .
[0042] (2) Use Convolution reduces the number of channels to C. m : ,in, This is the SiLU nonlinear activation function.
[0043] (3) Make three 1x1 projections onto V: , ,
[0044] Then, the space is expanded into the sequence length.
[0045]
[0046] (4) Calculate the similarity matrix and perform softmax:
[0047] (5) Weight the value with S and reshape the space: ,
[0048] (6) Use a 1x1 mapping to upgrade the channel to The output result is obtained through residual connection:
[0049] 3. Model Training Module: Configure training hyperparameters, set the number of training epochs to 100, batch size to 16, optimizer to AdamW, learning rate to 0.0001, and loss function to cross-entropy loss function. Train the ESCA-LCANet road environment perception model using the training dataset. In each iteration, calculate and update the weights of the ESCA-LCANet road environment perception model. At the end of each training epoch, use the validation set to evaluate the performance of the ESCA-LCANet road environment perception model, observe the accuracy and loss value on the validation set, and stop training when the model converges to prevent overfitting.
[0050] 4. Image Classification and Recognition Module: This module uses a test set to evaluate the classification and recognition performance of the ESCA-LCANet road environment perception model. It calculates the model's precision, accuracy, and F1 score on the test set. Based on the F1 score, number of parameters, computational cost, and precision, it constructs a visualization of the confusion matrix. This aims to systematically evaluate the model's classification performance and error distribution in real-world scenarios, and to assess the model's lightweight nature and required computational power. It also determines the adhesion effect for vehicles traveling on this type of road based on road environment conditions and provides detailed driving suggestions to improve driving safety.
[0051] 5. Risk Classification and Driving Guide Module: Constructs precise and quantifiable risk classifications and corresponding driving guidelines, transforming vague experiences into clear standards, as shown in the table below.
[0052] Table 1. Risk Classification and Driving Guidelines
[0053] A method for intelligent perception of road surface conditions based on image recognition includes the following steps: S1.1 (Constructing a dataset): Constructing a dedicated Road Environment Condition Dataset (RSCD) containing 21 typical scenarios, such as dry asphalt smoothness, wet asphalt damage, waterlogged concrete, and icy / snowy road surfaces. The dataset contains a total of 16,000 images, covering different weather and lighting conditions.
[0054] S1.2 (Data Partitioning and Augmentation): Divide the dataset into training, validation, and test sets in an 8:1:1 ratio. Preprocess the training set images, including: Filtering and denoising; data augmentation, using random rotation (within ±30°), random scaling (0.8-1.2 times), random horizontal / vertical flipping, and brightness adjustment (±20%) to increase data diversity and improve model generalization ability.
[0055] S2.1 (Model Architecture): Construct a road environment perception model named ESCA-LCAnet. Its innovation lies in: ESCA module: An enhancement module that combines efficient channel attention (ECA) with a spatial pyramid structure is embedded in the backbone network (such as ResNet-50) to enhance the model's ability to extract and focus on multi-scale road features.
[0056] LCAhead module: Design a lightweight classification head that reduces the number of parameters and computational complexity while maintaining accuracy.
[0057] Transfer learning is performed using pre-trained weights loaded onto large datasets such as ImageNet to accelerate convergence.
[0058] S2.2 (Model Training): Configure training hyperparameters, such as using the AdamW optimizer, setting the initial learning rate to 0.0001, employing cosine annealing, and setting the batch size to 16. Update model weights on the training set. Monitor performance using the validation set during training to prevent overfitting.
[0059] S3.1 (Model Evaluation): The trained ESCA-LCAnet model is evaluated using a test set, with accuracy, precision, recall, and F1 score as the main metrics to ensure that the model performance meets the application requirements.
[0060] S3.2 (State Classification and Risk Warning): Real-time acquired road images are input into the trained model, which outputs the recognition results of 21 road environment states. The system has a built-in risk classification database (for example, "dry asphalt smooth" is classified as low risk level 1, and "icy surface" is classified as the highest risk level 5), and is associated with specific driving guidelines (such as: "Slippery snowy road surface detected, high risk, please maintain low speed and avoid sharp turns and sudden braking").
[0061] Example 2: The computer-readable storage medium of this example stores a computer program that, when executed by a processor, implements the steps of the image recognition-based intelligent road condition perception method in Example 1.
[0062] The computer-readable storage medium in this embodiment can be an internal storage unit of the terminal, such as the terminal's hard disk or memory; the computer-readable storage medium in this embodiment can also be an external storage device of the terminal, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc. equipped on the terminal; furthermore, the computer-readable storage medium can include both the terminal's internal storage unit and external storage devices.
[0063] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.
[0064] Example 3: The computer device of this example includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the intelligent road condition perception method based on image recognition in Example 1.
[0065] In this embodiment, the processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The memory can include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.
[0066] Those skilled in the art will understand that the content disclosed in the embodiments can be provided as a method, system, or computer program product. Therefore, this solution can take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this solution can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage) containing computer-usable program code.
[0067] This solution is described with reference to flowchart illustrations and / or block diagrams of methods and computer program products according to embodiments of this solution. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a 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, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0068] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0069] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0070] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0071] The examples described herein are merely preferred embodiments of the invention and are not intended to limit the concept and scope of the invention. Any modifications and improvements made by those skilled in the art to the technical solutions of the invention without departing from the design concept of the invention should fall within the protection scope of the invention.
Claims
1. A road surface condition intelligent sensing system based on image recognition, characterized in that, include: The image preprocessing module is used to capture images and perform preprocessing operations, divide the training set, validation set and test set, and perform data augmentation operations; A road condition intelligent perception module that integrates the ESCA attention mechanism and LCAhead is based on the EfficientNetB0 model and integrates the ESCA attention mechanism and LCAhead module. Model training module; Image classification and recognition module; And a risk classification and driving guide module.
2. The system according to claim 1, characterized in that, The road condition intelligent perception module that integrates the ESCA attention mechanism and LCAhead consists of an EfficientNetB0 network model, which comprises convolutional layers, multiple MBConv modules, average pooling layers, max pooling layers, and softmax fully connected layers. The SE attention mechanism in the original MBConv module is replaced with the ESCA attention mechanism, and the last 1×1 convolutional module is replaced with the LCAhead module. While reducing parameters, it introduces lightweight spatial mixing, spatial self-attention mechanism, and residual reconstruction module to improve the ability to distinguish subtle road features.
3. The system according to claim 1, characterized in that, The model training module configures training hyperparameters, uses the cross-entropy loss function, trains the ESCA-LCANet road environment perception model using the training dataset, calculates and updates the weights of the ESCA-LCANet road environment perception model in each iteration, and at the end of each training round, uses the validation set to evaluate the performance of the ESCA-LCANet road environment perception model, observes the accuracy and loss value of the validation set, so that training stops when the model converges to prevent overfitting.
4. The system according to claim 1, characterized in that, The image classification and recognition module uses a test set to evaluate the classification and recognition performance of the ESCA-LCANet road environment perception model. It calculates the model's precision, accuracy, and F1 score on the test set. Based on the F1 score, number of parameters, computational cost, and precision, it constructs a visualization of the confusion matrix. The aim is to systematically evaluate the model's classification performance and error distribution in real-world scenarios, and to detect the model's lightweight nature and required computational power.
5. The system according to claim 1, characterized in that, The risk classification and driving guide module constructs a precise and quantifiable risk classification and corresponding driving guide, transforming vague experience into clear standards.
6. A method for intelligent perception of road surface conditions based on image recognition, characterized in that, Includes the following steps: Data cleaning was performed on the publicly available RSCD dataset, and road environment images with typical characteristics were selected to obtain a road environment perception dataset of 21 categories to be used. Data augmentation was performed on the data images. The acquired RSCD road environment status dataset was divided proportionally according to an 8:1:1 ratio to obtain a training dataset, a validation dataset, and a test dataset. Based on the EfficientNetB0 model, an improved ESCA-LCANet road environment perception model is constructed by integrating the ESCA attention mechanism and the LCAhead module. Configure the training hyperparameters and update the model weights based on the road environment perception dataset to obtain the trained ESCA-LCANet road environment perception model; The data is input into the ESCA-LCANet road environment perception model for training. The performance of the road environment perception dataset is evaluated and classified to identify different road environment states. The weights are then deployed to mobile devices to achieve real-time classification and recognition of road environment states.
7. The method according to claim 6, characterized in that, Data augmentation is applied to the data images, including random rotation, random scaling, random flipping, and brightness adjustment.
8. The method according to claim 6, characterized in that, The specific steps for constructing the improved ESCA-LCANet road environment perception model based on the EfficientNetB0 model, integrating the ESCA attention mechanism and the LCAhead module, are as follows: ESCA module: An enhancement module that combines efficient channel attention and spatial pyramid structure is embedded in the backbone network to enhance the model's ability to extract and focus on multi-scale road features; LCAhead module: Design a lightweight classification head that reduces the number of parameters and computational complexity while maintaining accuracy; Transfer learning is performed on pre-trained weights loaded onto large datasets to accelerate convergence; Configure training hyperparameters and update model weights on the training set; During training, performance is monitored using a validation set to prevent overfitting.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by the processor, it implements the steps of the image recognition-based intelligent perception method for road surface conditions as described in any one of claims 6-8.
10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the image recognition-based intelligent perception method for road surface conditions as described in any one of claims 6-8.