Plateau mountainous highway weather classification method and system based on dynamic fusion of image partition features

Abstract

The application relates to a highland mountain area highway weather classification method and system based on image partition feature dynamic fusion, and belongs to the technical field of traffic meteorology. The method performs road surface area identification on collected highway images with road surface areas to generate corresponding road surface area mask images; then, the highway images are divided into multiple subgraphs; the proportion of the road surface area in each subgraph is calculated according to the road surface area mask images; weather feature extraction is performed on each subgraph to calculate the probability distribution of various meteorological conditions of each subgraph; corresponding weight parameters are generated based on the probability distribution of various meteorological conditions of each subgraph and the road surface proportion information; the probability distribution of various meteorological conditions of each subgraph is weighted and fused by using the weight parameters to obtain the final meteorological condition classification result of the whole image. The application can effectively identify key meteorological factors that affect highway traffic safety in different scenes, thereby improving the traffic safety guarantee capability.

Description

Technical Field

[0001] This invention belongs to the field of traffic meteorology technology, specifically relating to a method and system for meteorological classification of highways in plateau and mountainous areas based on dynamic fusion of image partitioning features. Background Technology

[0002] Weather conditions on highways in high-altitude mountainous areas are highly variable, with frequent extreme weather events such as dense fog, torrential rain, and snow, seriously threatening traffic safety. Existing weather condition recognition technologies typically employ convolutional neural networks (CNNs) based on global images. These methods determine the weather category represented by an image by extracting features and classifying the entire image during training. Such methods usually rely on large-scale image datasets for training, failing to distinguish between road and background areas within the image, and directly analyzing and predicting the overall features of the entire image. For example, there is a weather image classification model based on the Xception architecture (Naufal, MF, & Kusuma, SF (2022). Weather image classification using convolutional neural network with transfer learning. AIP Conference Proceedings, 2470(1), 050004.). This study uses a CNN model based on the Xception architecture to classify weather images. There is also the MASK-CNN-Transformer (MASK-CT) model (Chen, S., Shu, T., Zhao, H., & Tang, YY (2023). MASK-CNN-Transformer for real-time multi-label weather recognition. Knowledge-Based Systems, 269, 110455.). This model combines convolutional neural networks (CNN) and Transformer architecture to propose a model for real-time multi-label weather recognition.

[0003] However, these methods have significant limitations. In images of high-altitude mountainous areas, distant mountains may exhibit meteorological phenomena unrelated to road traffic (such as dense fog or snow accumulation), and roadside water bodies may also lead to misclassification. These areas have limited impact on actual traffic safety. While these areas may exhibit abnormal weather phenomena, such as distant snow accumulation or mountain fog, they are usually far from the main road and have no direct relation to the actual road surface. If these areas are included as integral features in the identification process, it can easily lead to model misclassification, outputting weather categories that do not match the actual traffic environment. For example, if there is dense fog on distant mountains in an image, but the road surface is clear and unobstructed, traditional models may still misclassify the entire image as "dense fog weather," thus generating incorrect warnings. Traditional models fail to effectively distinguish the different contributions of road and non-road areas to the final classification result, leading to insufficient recognition accuracy in complex environments and making it difficult to provide reliable meteorological information support for traffic safety.

[0004] Furthermore, some existing studies have attempted to improve the model's ability to focus on key areas by introducing multi-scale feature extraction or attention mechanisms, but they still do not clearly guide the model to distinguish the importance difference between "road surface areas" and "non-road surface areas" in traffic safety judgments. In addition, most global image classification models lack explicit modeling of spatial structure and cannot adapt to the combined changes of complex terrain and meteorological backgrounds in plateau and mountainous areas, resulting in weak generalization ability in practical applications.

[0005] Therefore, overcoming the shortcomings of existing meteorological condition recognition technologies, such as their inability to effectively distinguish the different impacts of road and non-road areas on traffic safety, insufficient recognition accuracy in complex environments, difficulty in adapting to the combination of special terrain and meteorological changes in plateau and mountainous areas, and weak generalization ability, are problems that urgently need to be solved in this technical field. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies and provide a method and system for meteorological classification of highways in high-altitude mountainous areas by dynamically fusing image partitioning features.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for meteorological classification of highways in plateau and mountainous areas based on dynamic fusion of image partitioning features includes the following steps:

[0009] Step (1): The collected highway images with road surfaces are used to identify the road surface area and generate the corresponding road surface area mask map.

[0010] Step (2) divides the highway image into multiple sub-images and calculates the proportion of the road surface area in each sub-image based on the road surface area mask image;

[0011] Step (3): Extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map;

[0012] Step (4): Based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map, generate corresponding weight parameters;

[0013] Step (5) uses the weight parameters obtained in step (4) to weight and fuse the probability distributions of various meteorological conditions in each sub-image to obtain the final meteorological condition classification result of the whole image.

[0014] The core of the method disclosed in this invention lies in constructing a unified, end-to-end trainable intelligent classification model. Structurally, this model is divided into a "pre-training module" and a "training module." The "pre-training module" includes a road surface area recognition model that performs road surface mask generation and performs functions such as partitioning and proportion calculation; its parameters remain fixed during model training (i.e., parameter freezing). The "training module" integrates the core innovation of this invention: an improved visual encoder model for weather feature extraction and a dynamic weight calculation module; its parameters are jointly learned and updated during training based on the final classification loss. This end-to-end training mode with partially fixed parameters not only leverages existing mature models to improve processing efficiency but also ensures the collaborative operation of the core feature extraction and dynamic fusion mechanisms, enabling the model to efficiently learn and achieve accurate and robust classification of meteorological conditions on highways in plateau and mountainous areas.

[0015] Furthermore, preferably, in step (1), the specific method for identifying the road surface area of ​​the collected highway image with the road surface and generating the corresponding mask image is as follows: identify the collected highway image with the road surface, identify the road surface area and the non-road surface area, then set the pixel value of the road surface area to 1 and set the pixel value of the non-road surface area to 0, thereby obtaining the mask image.

[0016] Furthermore, preferably, for a highway image F containing road surfaces, the road surface mask m is represented as:

[0017] m = RoadSegModel(F)

[0018] Where RoadSegModel is the road surface region recognition model, and m is the generated binary mask image, satisfying:

[0019]

[0020] m(x,y) represents the pixel value at coordinates (x,y) on the mask image.

[0021] Furthermore, preferably, the specific method of step (2) is as follows:

[0022] Divide the highway image into N sub-images, denoted as f1 to f2. N For each sub-image, calculate its intersection with the road surface mask image m to obtain the proportion r of the road surface area in that sub-image. i , i = 1 to N; the calculation formula is:

[0023]

[0024] Where m(x,y) represents the pixel value at coordinates (x,y) on the mask image, 1 represents the road surface, and 0 represents the non-road surface; f i Let i be the i-th subgraph.

[0025] Furthermore, preferably, the specific method of step (3) is as follows: using an improved visual encoder model, for each sub-image f i The weather feature vector e is obtained by extracting the weather features. i Then, based on these features, the probability distribution y of meteorological conditions for each submap is calculated. i , i = 1 ~ N;

[0026] y i =[p i,1 ,p i,2 ... p i,K ], where p i,j This represents the probability that the i-th subgraph corresponds to the j-th weather condition, where j = 1 to K; K is the total number of weather categories;

[0027] y i The calculation method is as follows: The weather feature vector e... i The unnormalized category score Z is obtained by mapping it to the weather category through a fully connected layer. i Then, the probability distribution of weather conditions y is obtained by normalization using the Softmax function. i ;

[0028] Z i =w T e i +b

[0029] y i =Softmax(Z) i )

[0030] Where w is the initialized learning weight matrix, T represents the transpose operation, b is the initialized bias vector, and Softmax represents the normalized exponential function.

[0031] The improved visual encoder model includes a visual Transformer module, a dynamic feature fusion module, and an output module; the visual Transformer module, the dynamic feature fusion module, and the output module are connected sequentially.

[0032] The dynamic feature fusion module includes a spatial branch and a channel branch. The spatial branch captures local spatial details through 3×3 depthwise separable convolutions. The channel branch compresses the channel dimension and extracts global semantic information through sequentially connected 1×1 convolutions and SE channel attention mechanism. The number of channels after compression is C / r = 16, where C is the number of channels before compression and r is the compression ratio. The two branches extract features in parallel and then fuse them to output the weather feature vector.

[0033] Furthermore, preferably, K=18, the weather categories include no disaster weather, light rain, moderate rain, heavy rain, rainstorm, lightning, light fog, moderate fog, dense fog, light snow, moderate snow, heavy snow, blizzard, hail, sandstorm, halo, and icy roads; and the compression ratio r=16.

[0034] Furthermore, preferably, the specific method for generating the weight parameters in step (4) is as follows:

[0035] (a) For each subgraph i, determine its road surface area proportion r. i Probability distribution of weather conditions y i The features are concatenated to form an input feature vector x. i ;

[0036] (b) The input feature vector x i The input is fed into a fully connected neural network, which calculates the weight score s corresponding to the subgraph. i ;

[0037] (c) Weight scores s1 to s2 for all N subgraphs N The Softmax function is applied for normalization to obtain the final weight parameters for each subgraph, where the final weight parameter for subgraph i is w. i .

[0038] Furthermore, preferably, the specific method of step (5) is as follows:

[0039] The probability distribution of meteorological conditions of each sub-map is calculated by weighting and fusing the weight parameters obtained in step (4) using the following formula:

[0040]

[0041] Where y is a K-dimensional probability vector, y = [p1, p2, ..., p K ], p j This represents the probability that the entire image corresponds to the j-th weather condition, where K is the total number of weather categories;

[0042] Then, y is used as the final classification result of the meteorological conditions of the mountain roads in the entire image.

[0043] This invention also provides a meteorological classification system for highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features. The system employs the aforementioned method for meteorological classification of highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features, and includes:

[0044] The mask generation module is used to identify road surface areas in the acquired highway images containing road surfaces and generate corresponding road surface area mask maps.

[0045] The partitioning and proportion calculation module, connected to the mask generation module, is used to divide the highway image into multiple sub-images and calculate the proportion of the road surface area in each sub-image based on the road surface area mask image.

[0046] The weather feature extraction module, connected to the partitioning and proportion calculation module, is used to extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map.

[0047] The weight calculation module is connected to the partition and proportion calculation module and the weather feature extraction module, respectively. It is used to generate corresponding weight parameters based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map.

[0048] The weighted fusion module is connected to the weight calculation module and the weather feature extraction module, respectively. It is used to weight and fuse the probability distributions of various meteorological conditions of each sub-image using the weight parameters obtained by the weight calculation module, so as to obtain the final meteorological condition classification result of the whole image.

[0049] In this invention, RoadSegModel is a road surface region recognition model. As a general module, any artificial intelligence model with the ability to effectively recognize road surface regions in input road images can be applied to this method framework. For example, the RoadSegNet model using the DeepLabV3+ architecture (Pal, K., Yadav, P., & Katal, N. (2022). RoadSegNet: A deep learning framework for autonomous urban road detection. Journal of Engineering and Applied Science, 69(1), 1–21.), the road surface segmentation model based on the U-Net architecture (Spectrewolf8. (2024). Aerial Image Road Segmentation with U-Net-50 Architecture. Hugging Face.), and the road surface defect recognition model based on Mask R-CNN (Kulambayev, B., Kadyrov, T., Kydyrbek, A., et al.).

[0050] (2022).A Deep Learning-Based Approach for Road Surface Damage Detection.Computers,Materials&Continua,73(2),2715–2731.) These models can all provide effective road surface recognition results and generate mask maps suitable for subsequent processing in this method, thus demonstrating the versatility and flexibility of this module in the overall framework.

[0051] In this invention, the visual Transformer module is an existing model. This invention improves upon this model by adding a dynamic feature fusion module and an output module, thereby obtaining an improved visual encoder model.

[0052] The improved visual encoder model employs a dynamic feature fusion module to achieve multi-granularity feature extraction. This module includes a spatial branch and a channel branch. The spatial branch captures local spatial details through 3×3 depthwise separable convolutions, outputting C=256 channels with a stride of 2, achieving downsampling while preserving key spatial structures. The channel branch compresses the channel dimension (compression ratio r=16) using 1×1 convolutions and a SE channel attention mechanism, extracting global semantic information. After compression, the number of channels is C / r=16, reducing redundant information. The two branches extract features in parallel and then fuse them to obtain the weather feature vector.

[0053] In this invention, the computation module for generating dynamic weight parameters is designed to be lightweight and efficient, and is trained as part of the overall end-to-end model. The specific workflow of this module is as follows:

[0054] First, for each sub-map, the calculated road surface area percentage r is... i (A scalar) and the probability distribution of weather conditions y i (A K-dimensional vector) is concatenated to form a unified input feature vector x that can simultaneously represent road surface information and weather assessment. i .

[0055] Next, the input feature vector x i The data is fed into a lightweight fully connected neural network (FCNN). This network uses its internal learnable parameters to perform a non-linear transformation on the input features, thereby learning and determining the importance of the current subgraph to the final classification result, and outputting a weight score s. i .

[0056] Finally, the scores of all subgraphs are processed through a Softmax normalization layer. i This ensures that the sum of all weight parameters is 1.0, forming an effective probability distribution. The dynamically generated weight parameters w are thus... i It can adaptively fuse weather features from each sub-map, thereby improving the accuracy and reliability of the final classification.

[0057] During the training of the fully connected neural network of this invention, the proportion of the road surface area r is used as the reference. i Probability distribution of weather conditions y i As input, the weighted score s i As output, the goal is to minimize the loss (binary cross-entropy loss) between the weighted fusion of the probability distribution of weather conditions of each subgraph and the true label. In practical applications, a pre-trained fully connected neural network is used directly.

[0058] Classification loss

[0059] In this invention, y is a K-dimensional probability vector. The classification result is not just a single category, but rather the probability corresponding to each weather category. For example, if the weather category list is J = [heavy rain, heavy fog, heavy snow, ...] and y = [0.95, 0.85, 0.7, ...], then the classification result y represents that the probability of heavy rain is 95%, the probability of heavy fog is 85%, and the probability of heavy snow is 70% (i.e., multiple weather phenomena occur simultaneously).

[0060] This invention proposes a meteorological classification method for highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features. By introducing road surface mask-guided multi-sub-image feature extraction and a dynamic weight fusion mechanism, the accuracy and robustness of meteorological condition classification are significantly improved. The core innovation lies in the model's ability to learn a set of dynamic weight parameters based on the weather feature vector e of each sub-image. i The proportion of road surface area r i Probability distribution of weather conditions y i This method adaptively adjusts the weights of meteorological conditions in each sub-image and weights and fuses the probability distributions of various meteorological conditions in each sub-image to obtain the final meteorological condition classification result for the entire image. This approach not only effectively distinguishes the importance of road and non-road areas but also automatically adjusts the focus based on different scenarios. Therefore, it provides more reliable and accurate meteorological condition identification results under complex and variable meteorological conditions in high-altitude mountainous areas, offering strong technical support for traffic safety management.

[0061] This invention utilizes a dynamic fusion mechanism of image partitioning features to provide more accurate and reliable meteorological condition identification under complex weather conditions, especially in high-altitude mountainous areas. This invention can effectively identify key meteorological factors affecting road traffic safety in different scenarios, thereby improving traffic safety assurance capabilities.

[0062] Compared with the prior art, the beneficial effects of this invention are as follows:

[0063] 1. Improve classification accuracy and adaptive fusion

[0064] This invention significantly improves the classification accuracy of meteorological conditions by extracting and processing features from multiple subgraphs within the model and fusing them using learned dynamic weights. The model automatically learns the importance of different subgraphs to the final classification result, focusing on features of areas that truly impact traffic flow while reducing interference from background areas (such as distant snow-capped mountains and roadside water bodies). This adaptive fusion method effectively avoids judgment biases that may occur in traditional full-map recognition methods, ensuring accurate identification of key meteorological factors affecting traffic safety in special environments such as high-altitude mountainous areas.

[0065] 2. Enhance robustness

[0066] Because this invention employs a strategy of multi-subgraph feature processing and dynamic weight fusion within the model, it can effectively address potential local interference or abnormal regions in images. Even if the feature representation of some subgraphs is inaccurate (such as dense fog on distant mountains or reflections from roadside water), the model can automatically reduce the influence of these regions through the learned weight mechanism, thereby enhancing the robustness of the overall recognition. This mechanism makes the model more adaptable to various complex weather conditions and road environments, making it particularly suitable for complex environments with changeable weather, such as high-altitude mountainous areas.

[0067] 3. Wide range of applications

[0068] The dynamic weight fusion mechanism of this invention has strong generalization ability and can adapt to different types of highway scenarios and various weather conditions. Since the weight parameters are obtained through model learning rather than being manually set, it can automatically adapt to the characteristics of different scenarios and correctly identify the weather conditions that truly affect traffic safety. This adaptability makes this invention particularly suitable for special environments with variable climates and complex road conditions, such as high-altitude mountainous areas, providing more reliable technical support for traffic safety management.

[0069] 4. Technology Comparison

[0070] By comparing the method of this invention with existing advanced technologies in the field and conducting quantitative analysis, the data shows that the method of this invention has a significant ability to identify key meteorological factors in complex scenarios such as plateau and mountainous areas, and can provide more reliable technical support for traffic meteorological early warning systems.

[0071] The following is a comparison of the experimental performance (recognition accuracy) of various methods on the high-quality plateau mountain meteorological disaster scene image dataset built by this invention, as shown in Table 1. As can be seen from the table, although general models such as ConvNeXt, MaxViT, and InternImage have powerful performance, the method proposed in this invention, with its targeted "image partitioning" and "dynamic fusion" mechanisms, can more accurately focus on key road surface areas affecting traffic safety and effectively eliminate interference from the background environment. Therefore, under various complex meteorological conditions unique to plateau mountainous areas, the recognition accuracy has achieved stable and comprehensive superiority, especially in categories such as "fog," "snow," and "icy roads," which are highly dependent on road conditions. This fully demonstrates the advanced nature and practical value of this invention in the specialized application field of traffic meteorology.

[0072] Table 1

[0073] No severe weather 97.20％ 97.50％ 97.80％ 98.20％ Light rain 77.50％ 78.80％ 79.50％ 82.50％ Moderate rain 80.20％ 81.50％ 82.10％ 85.80％ heavy rain 84.10％ 84.70％ 85.50％ 89.10％ rainstorm 86.00％ 86.90％ 87.60％ 91.30％ lightning 90.10％ 90.80％ 91.50％ 93.70％ Light mist 83.50％ 84.20％ 85.40％ 90.60％ Medium fog 86.80％ 87.50％ 88.10％ 92.80％ Dense fog 89.90％ 90.60％ 91.20％ 95.30％ Dense fog 91.50％ 92.10％ 92.80％ 96.50％ Light Snow 72.80％ 74.20％ 75.50％ 81.70％ Moderate snow 76.50％ 77.60％ 78.30％ 84.60％ heavy snow 79.70％ 80.50％ 81.80％ 87.90％ Blizzard 84.00％ 84.90％ 85.60％ 90.20％ hail 78.20％ 79.10％ 80.30％ 85.10％ Sandstorm 92.00％ 92.50％ 93.20％ 95.90％ Halo 94.30％ 94.80％ 95.50％ 97.20％ icy roads 81.60％ 82.70％ 83.40％ 89.50％

[0074] Note: ConvNeXt, MaxViT, and InternImage are three advanced models in the field of image recognition in recent years. ConvNeXt, proposed by Zhuang Liu et al. in 2022, modernizes convolutional networks and achieves performance comparable to Transformer; MaxViT, proposed by Zhengzhong Tu et al. in 2022, constructs an efficient hybrid vision model by combining local convolution and global attention mechanisms; InternImage, proposed by Wenhai Wang et al. in 2023, explores large-scale vision fundamental models based on deformable convolution and performs excellently in multiple tasks. Attached Figure Description

[0075] Figure 1 This is a flowchart of the image partitioning feature dynamic fusion method for meteorological classification of highways in high-altitude mountainous areas according to the present invention.

[0076] Figure 2 This is an overall highway image in an example of the present invention;

[0077] Figure 3 This is a road surface area mask image of a highway in an example of the present invention;

[0078] Figure 4 This is a schematic diagram of dynamic fusion of image partitioning features in an example of the present invention;

[0079] Figure 5 This is the identification result of the partition feature fusion model in the example of the present invention;

[0080] Figure 6 The weather recognition result is shown in the example of the weather image classification model based on the Xception architecture in this invention.

[0081] Figure 7 The weather recognition result of the MASK-CT model in the example of this invention;

[0082] Figure 8 This is a schematic diagram illustrating the generation of the mask image in this invention;

[0083] Figure 9 This is a flowchart for obtaining the proportion of road surface area in each sub-graph according to the present invention.

[0084] Figure 10 This is a diagram of the improved visual encoder model architecture in this invention;

[0085] Figure 11 This is a diagram of the overall architecture of the end-to-end model in this invention. Detailed Implementation

[0086] The present invention will now be described in further detail with reference to the embodiments.

[0087] Those skilled in the art will understand that the following embodiments are for illustrative purposes only and should not be construed as limiting the scope of the invention. Where specific techniques or conditions are not specified in the embodiments, they are performed in accordance with the techniques or conditions described in the literature in the field or according to the product instructions. Materials or equipment whose manufacturers are not specified are all conventional products that can be obtained by purchase.

[0088] Example 1

[0089] A method for meteorological classification of highways in plateau and mountainous areas based on dynamic fusion of image partitioning features includes the following steps:

[0090] Step (1): The collected highway images with road surfaces are used to identify the road surface area and generate the corresponding road surface area mask map.

[0091] Step (2) divides the highway image into multiple sub-images and calculates the proportion of the road surface area in each sub-image based on the road surface area mask image;

[0092] Step (3): Extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map;

[0093] Step (4): Based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map, generate corresponding weight parameters;

[0094] Step (5) uses the weight parameters obtained in step (4) to weight and fuse the probability distributions of various meteorological conditions in each sub-image to obtain the final meteorological condition classification result of the whole image.

[0095] A meteorological classification system for highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features, comprising:

[0096] The mask generation module is used to identify road surface areas in the acquired highway images containing road surfaces and generate corresponding road surface area mask maps.

[0097] The partitioning and proportion calculation module, connected to the mask generation module, is used to divide the highway image into multiple sub-images and calculate the proportion of the road surface area in each sub-image based on the road surface area mask image.

[0098] The weather feature extraction module, connected to the partitioning and proportion calculation module, is used to extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map.

[0099] The weight calculation module is connected to the partition and proportion calculation module and the weather feature extraction module, respectively. It is used to generate corresponding weight parameters based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map.

[0100] The weighted fusion module is connected to the weight calculation module and the weather feature extraction module, respectively. It is used to weight and fuse the probability distributions of various meteorological conditions of each sub-image using the weight parameters obtained by the weight calculation module, so as to obtain the final meteorological condition classification result of the whole image.

[0101] Example 2

[0102] A method for meteorological classification of highways in plateau and mountainous areas based on dynamic fusion of image partitioning features includes the following steps:

[0103] Step (1): The collected highway images with road surfaces are used to identify the road surface area and generate the corresponding road surface area mask map.

[0104] Step (2) divides the highway image into multiple sub-images and calculates the proportion of the road surface area in each sub-image based on the road surface area mask image;

[0105] Step (3): Extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map;

[0106] Step (4): Based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map, generate corresponding weight parameters;

[0107] Step (5) uses the weight parameters obtained in step (4) to weight and fuse the probability distributions of various meteorological conditions in each sub-image to obtain the final meteorological condition classification result of the whole image.

[0108] In step (1), the specific method for identifying the road surface area and generating the corresponding mask image in the collected highway image with road surface is as follows: identify the collected highway image with road surface, identify the road surface area and non-road surface area, then set the pixel value of the road surface area to 1 and the pixel value of the non-road surface area to 0, thereby obtaining the mask image.

[0109] For a highway image F containing road surfaces, the road surface mask m is represented as:

[0110] m = RoadSegModel(F)

[0111] Where RoadSegModel is the road surface region recognition model, and m is the generated binary mask image, satisfying:

[0112]

[0113] m(x,y) represents the pixel value at coordinates (x,y) on the mask image.

[0114] The specific method for step (2) is as follows:

[0115] Divide the highway image into N sub-images, denoted as f1 to f2. N For each sub-image, calculate its intersection with the road surface mask image m to obtain the proportion r of the road surface area in that sub-image. i , i = 1 to N; the calculation formula is:

[0116]

[0117] Where m(x,y) represents the pixel value at coordinates (x,y) on the mask image, 1 represents the road surface, and 0 represents the non-road surface; f i Let i be the i-th subgraph.

[0118] The specific method of step (3) is as follows: using an improved visual encoder model, for each sub-image f i The weather feature vector e is obtained by extracting the weather features. i Then, based on these features, the probability distribution y of meteorological conditions for each submap is calculated. i , i = 1 ~ N;

[0119] y i =[p i,1 ,p i,2 ... p i,K ], where p i,j This represents the probability that the i-th subgraph corresponds to the j-th weather condition, where j = 1 to K; K is the total number of weather categories;

[0120] y i The calculation method is as follows: The weather feature vector e... iThe unnormalized category score Z is obtained by mapping it to the weather category through a fully connected layer. i Then, the probability distribution of weather conditions y is obtained by normalization using the Softmax function. i ;

[0121] Z i =w T e i +b

[0122] y i =Softmax(Z) i )

[0123] Where w is the initialized learning weight matrix, T represents the transpose operation, b is the initialized bias vector, and Softmax represents the normalized exponential function.

[0124] The improved visual encoder model includes a visual Transformer module, a dynamic feature fusion module, and an output module; the visual Transformer module, the dynamic feature fusion module, and the output module are connected sequentially.

[0125] The dynamic feature fusion module includes a spatial branch and a channel branch. The spatial branch captures local spatial details through 3×3 depthwise separable convolutions. The channel branch compresses the channel dimension and extracts global semantic information through sequentially connected 1×1 convolutions and SE channel attention mechanism. The number of channels after compression is C / r = 16, where C is the number of channels before compression and r is the compression ratio. The two branches extract features in parallel and then fuse them to output the weather feature vector.

[0126] K=18, weather categories include no disaster weather, light rain, moderate rain, heavy rain, rainstorm, lightning, light fog, moderate fog, dense fog, light snow, moderate snow, heavy snow, blizzard, hail, sandstorm, halo, and icy roads; compression ratio r=16.

[0127] In step (4), the specific method for generating the weight parameters is as follows:

[0128] (a) For each subgraph i, determine its road surface area proportion r. i Probability distribution of weather conditions y i The features are concatenated to form an input feature vector x. i ;

[0129] (b) The input feature vector x i The input is fed into a fully connected neural network, which calculates the weight score s corresponding to the subgraph. i ;

[0130] (c) Weight scores s1 to s2 for all N subgraphs NThe Softmax function is applied for normalization to obtain the final weight parameters for each subgraph, where the final weight parameter for subgraph i is w. i .

[0131] The specific method for step (5) is as follows:

[0132] The probability distribution of meteorological conditions of each sub-map is calculated by weighting and fusing the weight parameters obtained in step (4) using the following formula:

[0133]

[0134] Where y is a K-dimensional probability vector, y = [p1, p2, ..., p K ], p j This represents the probability that the entire image corresponds to the j-th weather condition, where K is the total number of weather categories;

[0135] Then, y is used as the final classification result of the meteorological conditions of the mountain roads in the entire image.

[0136] like Figure 11 As shown, a meteorological classification system for highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features is proposed. This system employs the aforementioned method for classifying meteorological conditions on highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features, and includes:

[0137] The mask generation module is used to identify road surface areas in the acquired highway images containing road surfaces and generate corresponding road surface area mask maps.

[0138] The partitioning and proportion calculation module, connected to the mask generation module, is used to divide the highway image into multiple sub-images and calculate the proportion of the road surface area in each sub-image based on the road surface area mask image.

[0139] The weather feature extraction module, connected to the partitioning and proportion calculation module, is used to extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map.

[0140] The weight calculation module is connected to the partition and proportion calculation module and the weather feature extraction module, respectively. It is used to generate corresponding weight parameters based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map.

[0141] The weighted fusion module is connected to the weight calculation module and the weather feature extraction module, respectively. It is used to weight and fuse the probability distributions of various meteorological conditions of each sub-image using the weight parameters obtained by the weight calculation module, so as to obtain the final meteorological condition classification result of the whole image.

[0142] Example 3

[0143] The core of the method disclosed in this invention lies in constructing a unified, end-to-end trainable intelligent classification model (i.e., an end-to-end model). The construction and workflow of this model can be decomposed into five steps as shown in the following embodiments. It should be noted that the "road area recognition model" described in step (1) can be regarded as an independent pre-training component in this invention, and its parameters are fixed during model training. The "improved visual encoder model" described in step (3) and the weight generation network described in step (4) together constitute the core module that needs to be learned. When performing end-to-end optimization on the overall model, the parameters of the "improved visual encoder model" and the parameters of the weight generation network will be learned and updated simultaneously according to the final classification loss, so that each step works together to achieve accurate and robust classification of the meteorological conditions of highways in plateau mountainous areas.

[0144] Figure 11 This paper details the overall architecture of the end-to-end model proposed in this invention, clearly dividing it into a "pre-training module" and a "training module." The parameters of the "pre-training module" are fixed (frozen), and it contains a "mask generation module" and a "partitioning and proportion calculation module." These two modules are responsible for performing common image processing steps, including generating road surface mask maps, segmenting the image into sub-images, and calculating the road surface proportion of each sub-image. The parameters of the "training module" are learnable and form the core of the entire model. It consists of a "weather feature extraction module," a "weight calculation module," and a "weighted fusion module" in that order. During training, the output of the pre-training module is fed into the training module. Through joint training, the parameters of its first two modules are optimized, enabling the model to learn to extract weather features from sub-images, adaptively calculate fusion weights, and ultimately output accurate classification results. This architectural design ensures the efficiency of the entire training process and the accuracy of the final classification.

[0145] A method for meteorological classification of highways in plateau and mountainous areas based on dynamic fusion of image partitioning features includes the following steps:

[0146] Step (1): The collected highway images with road surfaces are used to identify the road surface area and generate the corresponding road surface area mask map.

[0147] Step (2) divides the highway image into multiple sub-images and calculates the proportion of the road surface area in each sub-image based on the road surface area mask image;

[0148] Step (3): Extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map;

[0149] Step (4): Based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map, generate corresponding weight parameters;

[0150] Step (5) uses the weight parameters obtained in step (4) to weight and fuse the probability distributions of various meteorological conditions in each sub-image to obtain the final meteorological condition classification result of the whole image.

[0151] This invention provides a meteorological classification model for highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features. Unlike traditional global recognition methods, this invention first generates a road surface area mask map using the RoadSegModel, dividing the highway image into multiple sub-images. The proportion of the road surface area in each sub-image is calculated based on the road surface area mask map. An improved visual encoder model is then used to extract and process the features of these sub-image regions, obtaining meteorological condition feature representations for each sub-image region. Finally, a learnable dynamic weight fusion mechanism is used to fuse these sub-image feature representations into a meteorological condition classification result for the overall image. This fusion weight is not simply determined directly by the road surface proportion, but is learned by the model and can adaptively adjust according to the content features of the sub-images and their road surface proportion information. Figure 1 A flowchart of the method of the present invention is shown.

[0152] 1. Mask

[0153] Road surface masking is a key tool for identifying road surface regions in images. Each image generates a corresponding mask map using any pre-trained road surface region recognition model applicable to this method framework. The mask map identifies road surface and non-road surface regions in the image, providing a foundation for sub-image feature extraction and road surface proportion calculation within the model, and helping the model to pay closer attention to weather conditions directly related to road traffic. A schematic diagram of mask map generation is shown below. Figure 8 As shown.

[0154] For a captured highway image F containing the road surface, the road surface mask m can be represented as:

[0155] m = RoadSegModel(F)

[0156] Where RoadSegModel is the road surface region recognition model, and m is the generated binary mask image, satisfying:

[0157]

[0158] The mask image m not only directly affects the calculation of the road surface proportion in the sub-image, but also indirectly affects the subsequent feature extraction and weight generation process through the road surface proportion, making it an important foundation of the entire model.

[0159] 2. Sub-map division and road surface proportion calculation

[0160] The mask image is divided into N sub-regions, denoted as f1 to f2. NFor each sub-image, the proportion r of the road surface area in that sub-image is obtained by calculating its intersection with the road surface mask. i i = 1 to N, denoted as r1 to r N Flowchart as follows Figure 9 As shown. These road surface proportions are static parameters for the current input image, serving as additional input features for the model.

[0161] The road surface percentage for each submap can be expressed as:

[0162]

[0163] Where m(x,y) represents the pixel value at coordinates (x,y) on the mask image, 1 represents the road surface and 0 represents the non-road surface.

[0164] The flowchart for obtaining the proportion of road surface areas in each sub-graph is as follows: Figure 9 As shown.

[0165] 3. Extraction of weather features from multiple regions

[0166] The improved Vision Encoder model is based on the Visual Transformer (ViT). It captures semantic relationships between different regions in an image through a multi-head self-attention mechanism and incorporates a dynamic feature fusion module to achieve multi-granularity feature extraction, enhancing the ability to perceive local details, such as... Figure 10 As shown, the dynamic feature fusion module comprises a spatial branch and a channel branch. The spatial branch captures local spatial details through 3×3 depthwise separable convolutions, outputting C=256 channels with a stride of 2, achieving downsampling while preserving key spatial structures. The channel branch compresses the channel dimension (compression ratio r=16) through 1×1 convolutions and a SE channel attention mechanism, extracting global semantic information. After compression, the number of channels is C / r=16, reducing redundant information. The two branches extract features in parallel and then fuse them to obtain the weather feature vector. This model supports a multi-query attention mechanism, converting image planar data (32×32 resolution image patches) into a linear token sequence and predicting the next token through an autoregressive approach to obtain each sub-image f. i A unified embedding representation of weather features—weather feature vector e i Then, based on these features, the probability distribution y of meteorological conditions for each submap is calculated. i , i = 1 ~ N;

[0167] y i =[p i,1 ,p i,2 ... p i,K ], where p i,j, represents the probability of the i-th subgraph corresponding to the j-th weather condition, j = 1 to K; K is the total number of weather categories (including 18 typical high-altitude and mountainous weather conditions such as no disaster weather, light rain, moderate rain, heavy rain, rainstorm, lightning, light fog, moderate fog, dense fog, light snow, moderate snow, heavy snow, blizzard, hail, sandstorm, halo, and icy roads). For example, y1 represents the probability distribution of weather conditions obtained by the improved visual encoder model processing subgraph f1, which is a K-dimensional vector. i =[p i,1 ,p i,2 ... p i,K ], where p 1,j y1 represents the probability of the subgraph corresponding to the j-th weather condition, where K is the total number of weather categories; y2 represents the weather condition probability distribution obtained by processing subgraph f2 using the improved visual encoder model, which is also a K-dimensional vector. i =[p i,1 ,p i,2 ... p i,K ]; and so on up to y N .

[0168] The parameters of the improved vision encoder model are shown in Table 2:

[0169] Table 2

[0170]

[0171] 4. Dynamic weight generation and fusion mechanism

[0172] A core innovation of this invention lies in the dynamic weight generation mechanism, which is not a complex, independent network, but rather a learnable computational component within the end-to-end model. The weight parameters for each subgraph are generated using a lightweight, learnable dynamic weight calculation module. This module's input is only the calculated road surface area percentage r. i Probability distribution of weather conditions y i .

[0173] The specific calculation process is as follows:

[0174] First, for the i-th subgraph, determine its road surface area percentage (a scalar) r. i Its K-dimensional probability distribution vector of weather conditions y i Concatenate the vectors to form a (K+1) dimensional input feature vector x. i :

[0175] x i =concat(r i ,y i )

[0176] Subsequently, the feature vector x i The input is fed into a lightweight fully connected neural network (FCNN), which typically contains two hidden layers (32-dimensional). This network learns the non-linear relationship between input features and weight importance, and outputs a weight score s. i .

[0177] Finally, to ensure that the sum of the weights of all subgraphs is 1, thus forming an effective attention allocation mechanism, the model will collect the weight scores s of all N subgraphs. i Then, it is normalized using the Softmax function (the Softmax function assigns the highest-scoring subgraph the largest weight), resulting in the final weight parameters w used for weighted fusion. i :

[0178]

[0179] In this way, the model can automatically and learnably assign appropriate weights to each sub-map based on the road surface information and weather conditions, thereby highlighting the contribution of key areas in the final fusion step.

[0180] 5. Adaptive Fusion

[0181] The final classification result is obtained by weighted fusion of the probability distributions of meteorological conditions from each submap:

[0182]

[0183] Here, y is also a K-dimensional probability vector, representing the probability distribution of various weather conditions in the overall image. y = [p1, p2, ..., p K ], where p j y represents the probability that the entire image corresponds to the j-th type of weather condition, and K is the total number of weather categories; then y is used as the final classification result of the mountain road weather conditions for the entire image.

[0184] This fusion approach enables the proposed method to adaptively focus on the most representative areas that are relevant to traffic safety in different scenarios, thereby improving overall recognition accuracy.

[0185] like Figure 11 As shown, a meteorological classification system for highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features is proposed. This system employs the aforementioned method for classifying meteorological conditions on highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features, and includes:

[0186] The mask generation module is used to identify road surface areas in the acquired highway images containing road surfaces and generate corresponding road surface area mask maps.

[0187] The partitioning and proportion calculation module, connected to the mask generation module, is used to divide the highway image into multiple sub-images and calculate the proportion of the road surface area in each sub-image based on the road surface area mask image.

[0188] The weather feature extraction module, connected to the partitioning and proportion calculation module, is used to extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map.

[0189] The weight calculation module is connected to the partition and proportion calculation module and the weather feature extraction module, respectively. It is used to generate corresponding weight parameters based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map.

[0190] The weighted fusion module is connected to the weight calculation module and the weather feature extraction module, respectively. It is used to weight and fuse the probability distributions of various meteorological conditions of each sub-image using the weight parameters obtained by the weight calculation module, so as to obtain the final meteorological condition classification result of the whole image.

[0191] Application Examples

[0192] 1. Mask

[0193] First, the acquired highway image F containing the road surface is processed by any pre-trained road surface region recognition model applicable to this method framework to identify the road surface region in the image and generate a corresponding mask image m. Pixel values ​​of 1 in the mask image represent road surface regions, and 0 represents non-road surface regions.

[0194] 2. Sub-map division and road surface proportion calculation

[0195] In this example, the highway image F is divided into nine sub-images f1 to f9 in a 3x3 grid. For each sub-image, its intersection with the road surface mask image m is calculated to obtain the proportion of the road surface area r1 to r9 in that sub-image. These road surface proportion parameters will be used as part of the features and input into the model for subsequent calculations.

[0196] The calculation formula is:

[0197] 3. Multi-region feature extraction and processing

[0198] For each sub-image, the improved visual encoder model extracts its image features and calculates the probability distribution (y1, y2, ..., y9) of the corresponding weather conditions for that region. Each distribution y... iIt is a K-dimensional vector containing probability estimates of various possible weather categories in the corresponding subgraphs. In this embodiment, K=18, and these weather categories include 18 typical meteorological conditions in plateau and mountainous areas, such as no hazardous weather, light rain, moderate rain, heavy rain, rainstorm, lightning, light fog, moderate fog, dense fog, light snow, moderate snow, heavy snow, blizzard, hail, sandstorm, halo, and icy roads.

[0199] 4. Dynamic weight fusion

[0200] The learnable dynamic weight generation module can generate a set of dynamic weight parameters (w1, w2, ..., w9) based on the probability distribution of meteorological conditions in each sub-image and the proportion of road surface areas. These weight parameters are used to fuse the probability distributions of meteorological conditions in each sub-image to obtain the final classification result of the overall image.

[0201]

[0202] In this way, the model can adaptively adjust the importance of different subgraphs in the final decision, paying particular attention to areas that have a direct impact on traffic safety.

[0203] 5. Model Training

[0204] During the training phase, the method proposed in this invention learns feature extraction, sub-image meteorological condition probability distribution calculation, and dynamic weight generation processes in an end-to-end manner. By minimizing the classification loss function, the model can automatically learn an appropriate weight allocation strategy, enabling the final fusion result to accurately reflect the overall meteorological conditions of the image, especially key meteorological factors that have a practical impact on traffic safety.

[0205] The training objective can be expressed as:

[0206] Where θ represents the model parameters. Let y represent the classification loss function (such as binary cross-entropy loss). y is the model's final predicted output, which is the probability vector of weather conditions for the entire image obtained by weighted fusion of the probability distributions of each sub-image (each digit is a decimal between 0 and 1). It is a "multi-label probability distribution vector," where each element corresponds to the probability of a specific weather phenomenon occurring, thus allowing for judgments of multiple weather phenomena coexisting. true This is the true label corresponding to the image. It is a vector with the same dimension as y, and each bit in the vector corresponds to a "standard answer" (0 or 1) for a weather category.

[0207] The main parameters of the end-to-end model of this invention are shown in Table 3:

[0208] Table 3

[0209]

[0210]

[0211] Figures 2 to 5 The entire process of an example of the present invention is illustrated.

[0212] The acquired images are as follows Figure 2 As shown, then a mask is applied to obtain the following result: Figure 3 The mask diagram shown;

[0213] The highway image is then divided into multiple sub-images, and the proportion of the road surface area in each sub-image is calculated based on the road surface area mask image. Weather features are extracted for each sub-image, and the probability distribution of various weather conditions for each sub-image is calculated. Based on the probability distribution of various weather conditions and the road surface proportion information for each sub-image, corresponding weight parameters are generated, such as... Figure 4 As shown;

[0214] By weighting and fusing the probability distributions of various meteorological conditions from each sub-image using the obtained weight parameters, the final meteorological condition classification result for the entire image is obtained, such as... Figure 5 As shown, the identification results (visible light: no hazardous weather (confidence: 0.901), moderate fog (visibility 500-1000 meters) (confidence: 0.847)) accurately eliminated the interference of snowfall weather features outside the road surface area of ​​the highway image. Here, confidence level represents the probability of the weather condition.

[0215] Figure 6 and Figure 7 The images show the meteorological classification results for the same highway image under a non-regional global image weather recognition model. It can be noted that both global models identify snowfall, which is inconsistent with the actual traffic scenario (there is no snowfall in the road area).

[0216] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A meteorological classification method for highways in plateau and mountainous areas based on dynamic fusion of image partitioning features, characterized in that: Includes the following steps: Step (1): The collected highway images with road surfaces are used to identify the road surface area and generate the corresponding road surface area mask map. Step (2) divides the highway image into multiple sub-images and calculates the proportion of the road surface area in each sub-image based on the road surface area mask image; Step (3): Extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map; Step (4): Based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map, generate corresponding weight parameters; Step (5): The probability distributions of various meteorological conditions of each sub-image are weighted and fused using the weight parameters obtained in step (4) to obtain the final meteorological condition classification result of the whole image. The specific method of step (3) is as follows: using an improved visual encoder model, for each sub-image f i Weather feature vectors are obtained by extracting weather features. e i Then, based on these features, the probability distribution of meteorological conditions for each submap is calculated. y i , i =1~ N ; y i =[ p i,1 , p i,2 ,......, p i,K ],in, p i,j Indicates the first i The subgraph corresponds to the first The probability of similar weather conditions j =1~ K ; K Total number of weather categories; y i The calculation method is as follows: The weather feature vector... e i The unnormalized category score Z is obtained by mapping it to the weather category through a fully connected layer. i Then, the probability distribution of weather conditions is obtained by normalization using the Softmax function. y i ; in, w is the learning weight matrix obtained after initialization, T represents the transpose operation, b is the initial bias vector, and Softmax represents the normalized exponential function. The improved visual encoder model includes a visual Transformer module, a dynamic feature fusion module, and an output module; the visual Transformer module, the dynamic feature fusion module, and the output module are connected sequentially. The dynamic feature fusion module includes a spatial branch and a channel branch. The spatial branch captures local spatial details through 3×3 depthwise separable convolutions. The channel branch compresses the channel dimension and extracts global semantic information through sequentially connected 1×1 convolutions and SE channel attention mechanism. The number of channels after compression is C / r=16, where C is the number of channels before compression and r is the compression ratio. The two branches extract features in parallel and then fuse them to output the weather feature vector. In step (4), the specific method for generating the weight parameters is as follows: (a) For each subgraph i, determine the proportion of its road surface area. r i Probability distribution of weather conditions y i The features are concatenated to form an input feature vector. x i ; (b) The input feature vector x i The input is fed into a fully connected neural network, which calculates the weight score s corresponding to the subgraph. i ; (c) Weight scores s1 to s2 for all N subgraphs N The Softmax function is applied for normalization to obtain the final weight parameters for each subgraph, where the final weight parameters for subgraph i are: w i .

2. The image partitioning feature dynamic fusion method for meteorological classification of highways in plateau and mountainous areas according to claim 1, characterized in that, In step (1), the specific method for identifying the road surface area and generating the corresponding mask image in the collected highway image with road surface is as follows: identify the collected highway image with road surface, identify the road surface area and non-road surface area, then set the pixel value of the road surface area to 1 and the pixel value of the non-road surface area to 0, thereby obtaining the mask image.

3. The image partitioning feature dynamic fusion method for meteorological classification of highways in plateau and mountainous areas according to claim 2, characterized in that, For highway images with road surfaces Road surface mask Represented as: in, For road surface area identification model, The generated binary mask image satisfies: ； Represents coordinates on the mask image The pixel value at that location.

4. The image partitioning feature dynamic fusion method for meteorological classification of highways in plateau and mountainous areas according to claim 1, characterized in that, The specific method for step (2) is as follows: The highway image is divided into A subgraph, denoted as to For each subgraph, calculate its relationship with the road surface mask image. The intersection of these intersections yields the proportion of the road surface area in the sub-graph. r i , i =1~ N ; The calculation formula is: in, Represents coordinates on the mask image The pixel value at the location, 1 represents the road surface, and 0 represents the non-road surface; f i Let i be the i-th subgraph.

5. The image partitioning feature dynamic fusion method for meteorological classification of highways in plateau and mountainous areas according to claim 1, characterized in that, K =18, weather categories include no disaster weather, light rain, moderate rain, heavy rain, rainstorm, lightning, light fog, moderate fog, dense fog, light snow, moderate snow, heavy snow, blizzard, hail, sandstorm, halo, and icy roads; compression ratio r=16.

6. The image partitioning feature dynamic fusion method for meteorological classification of highways in plateau and mountainous areas according to claim 1, characterized in that, The specific method for step (5) is as follows: The probability distribution of meteorological conditions of each subplot is calculated by weighting and fusing the weight parameters obtained in step (4) using the following formula: in, For one K dimensional probability vector, , Indicates the entire image corresponds to the first... The probability of similar weather conditions K Total number of weather categories; Then with The final classification result of meteorological conditions on mountain roads in the whole image.

7. A meteorological classification system for highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features, employing the meteorological classification method for highways in high-altitude mountainous areas based on dynamic fusion of image partitioning features as described in any one of claims 1 to 6, characterized in that... include: The mask generation module is used to identify road surface areas in the acquired highway images containing road surfaces and generate corresponding road surface area mask maps. The partitioning and proportion calculation module, connected to the mask generation module, is used to divide the highway image into multiple sub-images and calculate the proportion of the road surface area in each sub-image based on the road surface area mask image. The weather feature extraction module, connected to the partitioning and proportion calculation module, is used to extract weather features for each sub-map and calculate the probability distribution of various meteorological conditions for each sub-map. The weight calculation module is connected to the partition and proportion calculation module and the weather feature extraction module, respectively. It is used to generate corresponding weight parameters based on the probability distribution of various meteorological conditions and road surface proportion information of each sub-map. The weighted fusion module is connected to the weight calculation module and the weather feature extraction module, respectively. It is used to weight and fuse the probability distributions of various meteorological conditions of each sub-image using the weight parameters obtained by the weight calculation module, so as to obtain the final meteorological condition classification result of the whole image.

Images