Road detection data analysis system

Abstract

The invention relates to the technical field of image processing, in particular to a road detection data analysis system which comprises an image definition enhancement module, a spectral analysis module, an enhancement real-time auxiliary module, a sign recognition and evaluation module, a sound recognition monitoring module, a weather condition influence analysis module, a road environment mapping module and a dynamic monitoring and anomaly detection module. According to the method, the image reconstruction quality is improved through the deep residual network resolution algorithm, the road material condition is accurately identified by using the hyperspectral imaging technology and the support vector machine algorithm, the maintenance efficiency is improved through the augmented reality technology, and the road sign identification accuracy is improved through the convolutional neural network model and the deep learning technology. The sound signal processing and the deep neural network improve the traffic abnormal event detection rate, the recurrent neural network model predicts the influence of the weather change on the road safety, the full convolutional network and the U-Net model accurately map the road environment, and the optical flow estimation method and the isolated forest algorithm monitor the road abnormal behavior in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a highway detection data analysis system. Background Art

[0002] The field of image processing technology involves the analysis, modification and improvement of image data to achieve specific goals, such as enhancing image quality, extracting important features, and identifying patterns and objects. This technical field is particularly critical in highway inspection data analysis systems because it can process highway images or video data captured by cameras and sensor devices. By applying various algorithms and technical means, such as edge detection, object recognition, and image segmentation, image processing aims to extract valuable information from raw data to support further analysis and decision making. Advances in this field, especially with the support of machine learning and artificial intelligence, have greatly improved the accuracy and efficiency of data processing, making it play a core role in highway inspection and analysis systems.

[0003] Among them, the highway detection data analysis system is an information system that integrates advanced image processing technology, which is designed to automatically collect, process and analyze highway image data to monitor and evaluate highway conditions. The main purpose of the system is to ensure highway safety, optimize traffic flow, identify maintenance needs in advance, and thus improve the efficiency and safety of road use. By analyzing the road surface, traffic sign visibility, and vehicle flow factors in real time or periodically, the system aims to provide a scientific basis for road maintenance, traffic planning, and safety management, thereby reducing traffic accidents, increasing road service life, and optimizing traffic management.

[0004] Traditional highway inspection data analysis systems, due to the lack of efficient image processing algorithms and detailed environmental perception capabilities, are unable to accurately identify and evaluate road surface damage, sign wear and environmental changes, resulting in slow response to road maintenance work and inaccurate maintenance decisions. Traditional systems also have obvious deficiencies in detecting abnormal traffic events and predicting the impact of weather conditions on roads, and are unable to effectively warn of potential safety risks, increasing the risk of traffic accidents. Summary of the invention

[0005] This application provides a highway inspection data analysis system to solve the problem that traditional highway inspection data analysis systems are unable to accurately identify and evaluate road surface damage, sign wear and environmental changes due to the lack of efficient image processing algorithms and detailed environmental perception capabilities, resulting in slow response to road maintenance work and inaccurate maintenance decisions. Traditional systems also have obvious deficiencies in detecting abnormal traffic events and predicting the impact of weather conditions on roads, and are unable to effectively warn of potential safety risks, increasing the risk of traffic accidents.

[0006] In view of the above problems, the present application provides a highway detection data analysis system.

[0007] The present application provides a highway detection data analysis system, wherein the system includes an image clarity enhancement module, a spectral analysis module, an enhanced real-time assistance module, a sign recognition and evaluation module, a sound recognition monitoring module, a weather condition impact analysis module, a road environment mapping module, and a dynamic monitoring and anomaly detection module;

[0008] The image definition enhancement module is based on the initial resolution road image and adopts a deep residual network resolution algorithm. It builds a deep residual block to learn the mapping relationship between resolution images, improves the image reconstruction quality by increasing the depth of the network, outputs a resolution image, identifies and analyzes the damage on the road surface, and generates a refined road image.

[0009] The spectral analysis module is based on the refined road image, adopts the hyperspectral imaging technology and the support vector machine algorithm, collects the multi-dimensional spectral data of the road surface to capture the characteristics of the reflected and absorbed light of the differentiated materials, uses the SVM to classify the multi-dimensional data, identifies the spectral signatures of the differentiated materials, monitors the road wear and oil pollution, and generates the material condition analysis results;

[0010] The enhanced real-time assistance module uses augmented reality technology based on the material condition analysis results, combines real-time road damage data with the user's field of view, and uses mobile devices and AR glasses to overlay and display virtual information, assisting maintenance personnel in locating the damage location and providing maintenance solutions, thereby improving maintenance efficiency and accuracy and generating an enhanced real-time maintenance assistance view;

[0011] The sign recognition and assessment module is based on the enhanced real-time maintenance auxiliary view and adopts a convolutional neural network model to identify road signs from images through deep learning technology, and assess their wear and occlusion status, thereby improving the accuracy of sign recognition, assessing the visibility and integrity of the signs, and generating a sign status assessment result;

[0012] The sound recognition monitoring module uses sound signal processing technology and deep neural network based on the sign status evaluation results to identify the target sound pattern by performing frequency analysis on the sound signal and training the deep learning model, thereby improving the detection rate of traffic anomaly events and generating traffic anomaly monitoring results;

[0013] The weather condition impact analysis module is based on the traffic anomaly monitoring results and adopts a recurrent neural network model to learn the impact of weather condition changes on road conditions by analyzing the time information in the sequence images, and comprehensively processes and analyzes historical and real-time weather data through a time series analysis method to predict the potential impact of weather changes on road safety and generate weather condition impact results;

[0014] The road environment mapping module uses a full convolutional network and U-Net model to classify images at the pixel level based on the results of weather conditions, distinguish between roads, vehicles, and pedestrians, and improve the accuracy of environmental recognition through image segmentation technology to generate a road environment map;

[0015] The dynamic monitoring and anomaly detection module is based on the road environment map, adopts the optical flow estimation method and the isolation forest algorithm, evaluates the movement direction and speed of the object by analyzing the pixel movement between consecutive image frames, and uses the isolation forest algorithm to detect abnormal behavior. It combines the dynamic monitoring and anomaly detection methods to monitor road usage in real time and identify abnormal behavior, and generate real-time monitoring and anomaly notification results.

[0016] Preferably, the refined road image includes road surface damage features, road markings, and road environment details; the material condition analysis results include road wear degree, oil distribution, and spectral signatures of differentiated materials; the enhanced real-time maintenance auxiliary view specifically combines road damage location, maintenance plan, and virtual auxiliary information; the sign status assessment results include recognition information of road signs, wear degree assessment, and occlusion status; the traffic anomaly monitoring results include target sound pattern recognition and abnormal event detection; the weather condition impact results include an assessment of the impact of weather changes on road conditions and a prediction of the potential impact of weather conditions on road safety in future time periods; the road environment map includes environmental elements that distinguish roads, vehicles, and pedestrians; the real-time monitoring and abnormal notification results include the direction of movement of objects, speed assessment, and detection of abnormal behavior.

[0017] Preferably, the image definition enhancement module includes an image input submodule, a resolution processing submodule, and a refined image output submodule;

[0018] The image input submodule loads image data based on the initial resolution road image, uses the PIL library to read the image, adjusts the image size to 256x256 pixels, adjusts the format to meet the input requirements of the deep learning model, and generates a processed road image;

[0019] The resolution processing submodule uses a deep residual network to perform resolution enhancement based on the processed road image, uses Python TensorFlow to build the network, sets the residual block to include multiple 3x3 convolutional layers, and uses a linear rectification function as the activation function. By adding multiple residual blocks, the image resolution is gradually improved to generate an enhanced resolution reconstructed image.

[0020] The refined image output submodule reconstructs the image based on the enhanced resolution, uses the Gaussian blur algorithm of the open source computer vision library for smoothing, reduces image noise, retains key texture information, applies the Canny edge detection algorithm to highlight the information of the damaged area, and enhances the visibility of the damage by identifying the boundary of the damaged area in the image to generate a refined road image.

[0021] Preferably, the spectrum analysis module includes a spectrum image acquisition submodule, a spectrum feature analysis submodule, and a material condition identification submodule;

[0022] The spectral image acquisition submodule collects spectral data based on the refined road image, uses a hyperspectral camera to capture the spectral characteristics of the road surface in the wavelength range of 400-1000nm, obtains the reflection and absorption spectral information of the road surface material, analyzes the road conditions, and generates a spectral data image;

[0023] The spectral feature analysis submodule extracts spectral features based on the spectral data image, uses principal component analysis to extract key features of the spectral data, and the principal component analysis converter sets the number of principal components to 10, retains key variation information in the spectral data, and generates spectral feature information;

[0024] The material condition identification submodule identifies the material condition based on the spectral feature information, uses a support vector machine, builds a support vector machine model in the scikit-learn library, sets the parameter C to 1.0, selects RBF as the kernel function, identifies road wear and oil stains, and generates material condition analysis results.

[0025] Preferably, the enhanced real-time auxiliary module includes a damage data integration submodule, an AR view generation submodule, and a maintenance auxiliary display submodule;

[0026] The damage data integration submodule collects and integrates road damage data based on the material condition analysis results, and uses the DataFrame in the Pandas library to organize the data, including screening key damage features including type, location, and degree, applying data standardization processing to unify the data format, and generating comprehensive road damage data;

[0027] The AR view generation submodule creates an augmented reality view based on the comprehensive road damage data, uses Unity3D combined with the Vuforia software development kit to build an AR scene, dynamically binds the real-time road damage data to the virtual object through the scripting language, adjusts the size, position, and color of the virtual object to match the real-time damage situation, performs synchronous updates of the three-dimensional interactive view, and generates a real-time AR maintenance view;

[0028] The maintenance auxiliary display submodule performs auxiliary display of maintenance plans based on the real-time AR maintenance view, uses OpenCV for image processing, identifies the edges and shapes of damaged areas, and combines virtual markers in the AR view to provide damage location and maintenance direction guidance, thereby generating an enhanced real-time maintenance auxiliary view.

[0029] Preferably, the sign recognition and evaluation module includes an image recognition submodule, a state evaluation submodule, and an evaluation result output submodule;

[0030] The image recognition submodule performs image recognition of road signs based on the enhanced real-time maintenance auxiliary view, builds a convolutional neural network model using the Python TensorFlow library, defines a model structure including multiple Conv2D layers and MaxPooling2D layers to extract image features, sets the optimizer to Adam and the loss function to categorical_crossentropy, performs model training and verification, and generates road sign recognition information;

[0031] The state assessment submodule assesses the wear and occlusion state of the road sign based on the road sign recognition information, uses an image processing algorithm, applies the Sobel operator to perform edge detection, and uses the HSV color space conversion to analyze the color saturation and brightness of the sign, assesses the visibility and integrity of the sign, and generates a road sign integrity analysis result;

[0032] The evaluation result output submodule outputs the final evaluation result based on the road sign integrity analysis result, uses the Python Matplotlib library for data visualization, integrates the evaluation indicators, forms an evaluation chart, provides a comprehensive description of the sign status, and generates the sign status evaluation result.

[0033] Preferably, the sound recognition monitoring module includes a sound data collection submodule, a sound event recognition submodule, and an abnormal event result submodule;

[0034] The sound data acquisition submodule collects sound data based on the sign status evaluation result, applies Butterworth filter to remove background noise in the field of digital signal processing through sound signal processing technology, and uses Fourier transform to extract the spectrum characteristics of the sound signal, so that the data quality meets the subsequent processing requirements, and generates processed sound data;

[0035] The sound event recognition submodule performs sound event recognition based on the processed sound data, applies a deep learning algorithm to build a deep neural network through Python and TensorFlow library, sets a network structure including multiple convolutional layers Conv2D for extracting time-frequency features of sound signals, processes the timing information of the sound through a recurrent layer, and recognizes the target sound patterns of vehicle horns and emergency vehicle alarms through training, and generates sound event recognition results;

[0036] The abnormal event result submodule is based on the sound event recognition results, adopts the data aggregation method, uses Python to integrate and analyze the recognized sound events, and identifies and evaluates traffic anomalies, including sudden accidents and emergency braking sounds, determines abnormal changes in traffic flow, and generates traffic anomaly monitoring results.

[0037] Preferably, the weather condition impact analysis module includes an image sequence processing submodule, a weather impact analysis submodule, and an impact result generation submodule;

[0038] The image sequence processing submodule performs image sequence processing based on traffic anomaly monitoring results, adopts a recurrent neural network, and builds an RNN model through the Python Keras library. The model uses a recurrent layer including LSTM to process image sequences, analyze the time change information in the sequence images, learn the impact of weather condition changes on road conditions, and generate time series image analysis results;

[0039] The weather impact analysis submodule performs weather condition impact analysis based on the time series image analysis results, uses time series analysis technology, including autoregressive models, combines historical and real-time weather data, performs historical data trend analysis and real-time data response analysis through Python, predicts the potential impact of weather changes on road safety, and generates road safety meteorological forecast results;

[0040] The impact result generation submodule generates impact results based on road safety meteorological forecast results, uses Python and data visualization tools including Matplotlib to display the impact of weather changes on road conditions, integrates analysis results, provides a display of the impact of weather changes on road safety, and generates weather condition impact results.

[0041] Preferably, the road environment mapping module includes an image segmentation submodule, an environment recognition submodule, and a map updating submodule;

[0042] The image segmentation submodule performs image segmentation based on the results of weather conditions. It uses a fully convolutional network and uses the Python TensorFlow library to build a network. The network includes a convolution layer for extracting image features and an upsampling layer for increasing image resolution. It performs pixel-level classification on the image, distinguishes environmental elements such as roads, vehicles, and pedestrians, and generates image segmentation results.

[0043] The environment recognition submodule uses the U-Net model for environment recognition based on the image segmentation results. The U-Net model is built using the Keras library, and the image segmentation accuracy is improved by splicing feature maps. Differentiated elements in the road environment, including roads, vehicles, and pedestrians, are distinguished and recognized to generate environment recognition results.

[0044] Based on the environmental recognition results, the map update submodule adopts geographic information system technology to integrate the recognized environmental elements, including roads, vehicles, and pedestrians, into the map data, update and optimize the spatial data, reflect the real-time road environment conditions, and generate a road environment map.

[0045] Preferably, the dynamic monitoring and anomaly detection module includes a dynamic data acquisition submodule, a real-time monitoring and analysis submodule, and an anomaly identification and notification submodule;

[0046] The dynamic data acquisition submodule collects dynamic road data based on the road environment map, applies the optical flow estimation method, uses the pyramid Lucas-Kanade function in the Python computer vision library, analyzes the movement of pixels between consecutive image frames, evaluates the movement direction and speed of the object, and generates dynamic road data;

[0047] The real-time monitoring and analysis submodule uses the isolation forest algorithm to detect abnormal behavior based on dynamic road data, sets the parameters of the isolation forest model, including the number of trees and the number of samples, uses the isolation forest function in the Python scientific computing library, identifies abnormal points in the data, and generates abnormal behavior recognition results;

[0048] The anomaly identification and notification submodule integrates abnormal behavior data based on the abnormal behavior identification results, uses data aggregation technology to summarize abnormal information, forms real-time monitoring results, sends alarm notifications through the communication interface, reminds related personnel to respond to emergencies, and generates real-time monitoring and abnormal notification results.

[0049] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0050] The image reconstruction quality is improved through the deep residual network resolution algorithm, the road material condition is accurately identified using hyperspectral imaging technology and support vector machine algorithm, maintenance efficiency is improved using augmented reality technology, the road sign recognition accuracy is improved using convolutional neural network model and deep learning technology, the detection rate of abnormal traffic events is improved using sound signal processing and deep neural network, the impact of weather changes on road safety is predicted using recurrent neural network model, the road environment is accurately mapped using fully convolutional network and U-Net model, and abnormal road behavior is monitored in real time using optical flow estimation method and isolation forest algorithm.

[0051] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A module diagram of a highway detection data analysis system is provided for the present invention;

[0053] Figure 2 The system framework diagram of the highway detection data analysis system proposed by the present invention;

[0054] Figure 3 A schematic diagram of a specific process of an image definition enhancement module of a highway detection data analysis system proposed by the present invention;

[0055] Figure 4 A specific flow chart of the spectrum analysis module of the highway detection data analysis system proposed by the present invention;

[0056] Figure 5 A specific flow chart of an enhanced real-time auxiliary module of a highway detection data analysis system proposed by the present invention;

[0057] Figure 6 A specific flow chart of the sign recognition and evaluation module of the highway detection data analysis system proposed by the present invention;

[0058] Figure 7 A specific flow chart of a sound recognition monitoring module of a highway detection data analysis system proposed by the present invention;

[0059] Figure 8 A specific flow chart of a weather condition impact analysis module of a highway detection data analysis system proposed by the present invention;

[0060] Figure 9 A specific flow chart of a road environment mapping module of a highway detection data analysis system proposed by the present invention;

[0061] Figure 10The present invention proposes a specific flow chart of the dynamic monitoring and anomaly detection module of the highway detection data analysis system. DETAILED DESCRIPTION

[0062] This application provides a highway inspection data analysis system to solve the problem that traditional highway inspection data analysis systems are unable to accurately identify and evaluate road surface damage, sign wear and environmental changes due to the lack of efficient image processing algorithms and detailed environmental perception capabilities, resulting in slow response to road maintenance work and inaccurate maintenance decisions. Traditional systems also have obvious deficiencies in detecting abnormal traffic events and predicting the impact of weather conditions on roads, and are unable to effectively warn of potential safety risks, increasing the risk of traffic accidents.

[0063] Application Overview

[0064] In the existing technology, traditional highway inspection data analysis systems are unable to accurately identify and evaluate road surface damage, sign wear and environmental changes due to the lack of efficient image processing algorithms and detailed environmental perception capabilities, resulting in slow response to road maintenance work and inaccurate maintenance decisions. Traditional systems also have obvious deficiencies in detecting abnormal traffic events and predicting the impact of weather conditions on roads, and are unable to effectively warn of potential safety risks, increasing the risk of traffic accidents.

[0065] In response to the above technical problems, the overall idea of ​​the technical solution provided by this application is as follows:

[0066] like Figure 1 , 2 As shown, the present application provides a highway detection data analysis system, wherein the system includes an image clarity enhancement module, a spectral analysis module, an enhanced real-time auxiliary module, a sign recognition and evaluation module, a sound recognition monitoring module, a weather condition impact analysis module, a road environment mapping module, and a dynamic monitoring and anomaly detection module;

[0067] The image clarity enhancement module is based on the initial resolution road image and adopts the deep residual network resolution algorithm. It builds a deep residual block to learn the mapping relationship between resolution images, improves the image reconstruction quality by increasing the depth of the network, outputs the resolution image, identifies and analyzes the damage on the road surface, and generates a refined road image.

[0068] The spectral analysis module is based on the refined road image, using hyperspectral imaging technology and support vector machine algorithm. It collects multi-dimensional spectral data of the road surface to capture the characteristics of reflected and absorbed light of differentiated materials, uses SVM to classify and process multi-dimensional data, identifies the spectral signatures of differentiated materials, monitors road wear and oil pollution, and generates material condition analysis results.

[0069] The enhanced real-time assistance module uses augmented reality technology based on the material condition analysis results. By combining real-time road damage data with the user's field of view, the module uses mobile devices and AR glasses to overlay virtual information for display, assisting maintenance personnel in locating the damage and providing maintenance solutions, improving maintenance efficiency and accuracy, and generating an enhanced real-time maintenance assistance view.

[0070] The sign recognition and assessment module is based on the enhanced real-time maintenance assistance view and adopts a convolutional neural network model. It uses deep learning technology to identify road signs from images and assess their wear and occlusion status, improve the accuracy of sign recognition, assess the visibility and integrity of signs, and generate sign status assessment results.

[0071] The sound recognition monitoring module uses sound signal processing technology and deep neural networks based on the sign status assessment results. It identifies the target sound pattern by performing frequency analysis on the sound signal and training the deep learning model, thereby improving the detection rate of traffic anomalies and generating traffic anomaly monitoring results.

[0072] The weather condition impact analysis module is based on traffic anomaly monitoring results and adopts a recurrent neural network model to learn the impact of weather condition changes on road conditions by analyzing the time information in the sequence images. It also uses a time series analysis method to comprehensively process and analyze historical and real-time weather data, predict the potential impact of weather changes on road safety, and generate weather condition impact results.

[0073] The road environment mapping module uses a fully convolutional network and U-Net model to classify images at the pixel level based on the results of weather conditions, distinguish between roads, vehicles, and pedestrians, and improve the accuracy of environmental recognition through image segmentation technology to generate a road environment map.

[0074] The dynamic monitoring and anomaly detection module is based on the road environment map and adopts the optical flow estimation method and the isolation forest algorithm. It evaluates the movement direction and speed of the object by analyzing the pixel movement between consecutive image frames, and uses the isolation forest algorithm to detect abnormal behavior. It combines the dynamic monitoring and anomaly detection methods to monitor road usage in real time and identify abnormal behavior, and generate real-time monitoring and anomaly notification results.

[0075] The refined road image includes road surface damage features, road markings, and road environment details. The material condition analysis results include the degree of road wear, the distribution of oil stains, and the spectral signatures of different materials. The enhanced real-time maintenance assistance view specifically combines the road damage location, the maintenance plan, and virtual assistance information. The sign status assessment results include the identification information of road signs, the assessment of wear degree, and the occlusion status. The traffic anomaly monitoring results include the recognition of target sound patterns and the detection of abnormal events. The weather condition impact results include the assessment of the impact of weather changes on road conditions and the prediction of the potential impact of future weather conditions on road safety. The road environment map includes environmental elements that distinguish roads, vehicles, and pedestrians. The real-time monitoring and anomaly notification results include the movement direction of objects, the speed assessment, and the detection of abnormal behaviors.

[0076] In the image sharpness enhancement module, the road image with the initial resolution is processed by a Deep Residual Network (DRN). The core of the DRN lies in its deep residual blocks, which capture the complex non-linear relationships between images to map the differences between high-resolution and low-resolution images. The residual blocks contain multiple convolutional layers, and each layer uses specific filters and activation functions. For example, 3x3 filters and the ReLU activation function are used to extract features and perform non-linear transformations. The increase in network depth is achieved by adding more residual blocks, and each block is connected by skip connections to ensure the unobstructed transmission of information flow in the network and avoid the vanishing gradient problem in the training of deep networks. After the network training is completed, the low-resolution road image is input, and the DRN extracts features and maps it, finally outputting a road image with higher sharpness. The output image is used to identify and analyze road surface damage, generate a refined road image, and effectively improve the analysis ability and accuracy of the road detection system.

[0077] In the spectral analysis module, hyperspectral imaging technology and the Support Vector Machine (SVM) algorithm are adopted to process based on the refined road image. Hyperspectral imaging technology captures the reflection and absorption characteristics of light by different materials by collecting multi-dimensional spectral data on the road surface. The spectral data is converted into multi-dimensional feature vectors as the input for the SVM algorithm. The SVM algorithm classifies the data in multi-dimensional space by constructing one or more hyperplanes. The core lies in selecting appropriate kernel functions (such as the radial basis function) and parameter settings (such as the penalty parameter C and the kernel parameter gamma) to optimize the classification accuracy. Through the training process, the SVM model learns the spectral signatures of different materials, classifies the captured spectral data, and identifies road wear and oil stain conditions. This module outputs the material condition analysis results, providing accurate data support for road maintenance and repair, and effectively improving the efficiency and quality of road maintenance.

[0078] In the enhanced real-time assistance module, augmented reality technology is used to provide real-time maintenance assistance based on the results of material condition analysis. The module first integrates road damage data and user vision to create an interactive visual interface through mobile devices and AR glasses. Augmented reality technology displays the damage location and related information in real time by superimposing virtual information on the actual visual scene. The system needs to process real-time damage data, synchronize it with the user's geographic location and perspective, and present it on the AR device through graphics rendering technology. Not only accurate spatial positioning technology is required, but also efficient image processing and rendering algorithms are required to ensure real-time performance. In this way, maintenance personnel can accurately locate the damage location and obtain specific maintenance plan recommendations. The enhanced real-time maintenance assistance view finally generated by this module improves maintenance efficiency and accuracy, bringing substantial improvements to road maintenance work.

[0079] In the sign recognition and evaluation module, the system uses deep learning technology based on convolutional neural networks (CNN) to recognize and evaluate the status of road signs in the enhanced real-time maintenance auxiliary view. The CNN model is composed of multiple convolutional layers, pooling layers, and fully connected layers. Each convolutional layer uses a set of learned filters to extract local features of the image. The pooling layer is used to reduce the feature dimension and enhance the robustness of the features. During the training phase, the model learns the features of the signs, such as shape, color, and symbols, through a large number of labeled road sign images. After training, the model can recognize road signs from the real-time maintenance auxiliary view and evaluate their wear and occlusion status. High-accuracy sign recognition is achieved through deep learning technology, and the visibility and integrity of the signs are comprehensively evaluated. The sign status evaluation results generated by this module provide important decision-making support for road safety management, ensuring the clear visibility and effective guidance of road signs.

[0080] In the sound recognition monitoring module, the system collects environmental sound signals as input data, which are preprocessed, including denoising and normalization, to ensure the accuracy of the analysis. Fast Fourier transform (FFT) is used to perform frequency analysis on sound signals, and the sound signals are converted into frequency domain representation to facilitate the identification of different frequency components. Deep neural networks, especially convolutional neural networks (CNN) and long short-term memory networks (LSTM), are used to process these frequency domain data and learn the complex characteristics of sound patterns. CNN is responsible for extracting local features in sound signals, while LSTM processes the time series relationship of these features to identify specific sound patterns, such as traffic accident sounds or emergency vehicle sounds. The training process involves a large number of labeled sound samples. The network weights are optimized through the back propagation algorithm to improve the recognition accuracy of the model. After the model training is completed, it can process sound data in real time, identify and classify target sound events, and generate traffic anomaly monitoring results. It not only improves the detection rate of traffic anomalies, but also reduces the probability of false alarms by accurately identifying sound patterns, thereby improving the overall efficiency of the traffic monitoring system.

[0081] In the weather condition impact analysis module, the system integrates a recurrent neural network (RNN), especially its variants such as long short-term memory network (LSTM) or gated recurrent unit (GRU), to process the time information in the sequence image and the historical and real-time weather data. The module receives the traffic anomaly monitoring results and weather data as input. The weather data includes temperature, humidity, precipitation, and is in the format of time series. Through the in-depth analysis of the sequence image and weather data, the model learns the relationship between weather changes and road conditions. The application of LSTM or GRU allows the system to remember long-term dependencies and effectively handle the long-distance dependency problem of time series data. The time series analysis method further processes these data comprehensively, including smoothing and trend prediction, to identify the potential impact of weather condition changes on road safety. Through this in-depth analysis, the system is able to predict road conditions under different weather conditions and generate weather condition impact results, which are important for guiding traffic management decisions and warning drivers in advance to avoid driving in bad weather conditions.

[0082] In the road environment mapping module, the system uses a fully convolutional network (FCN) and U-Net model to analyze the collected road images. These images are high-resolution RGB images. FCN is used to achieve pixel-level classification in the image, and can identify different elements in the image, such as roads, vehicles, pedestrians, etc. The U-Net model enhances the accuracy of image segmentation, especially in the processing of details on the edges of the image. The combined use of these two models enables the system to accurately distinguish and identify different components in the road environment by learning a large amount of labeled image data. The application of image segmentation technology not only improves the accuracy of environmental recognition, but also provides accurate basic data for dynamic monitoring and anomaly detection. The generated road environment map provides detailed road usage and environmental conditions information for subsequent modules, laying the foundation for real-time monitoring and anomaly detection.

[0083] In the dynamic monitoring and anomaly detection module, the system uses the optical flow estimation method and the isolation forest algorithm to analyze continuous image frames. The optical flow estimation method is used to calculate the movement of pixels between continuous frames to evaluate the movement direction and speed of the object. Based on the time change of pixel intensity, the movement characteristics of the object are identified by calculating the movement amount of pixels between continuous frames. The isolation forest algorithm is used to detect abnormal behavior. It constructs a decision tree by randomly selecting features and randomly splitting values ​​to isolate abnormal points. This method is particularly effective for detecting low-frequency abnormal behaviors because abnormal behaviors are relatively rare in data sets. Through the combination of these technologies, the system can monitor road usage in real time, identify abnormal behaviors including illegal lane changes and sudden parking, and generate real-time monitoring and abnormal notification results, which greatly improves the real-time and accuracy of road safety monitoring and provides effective decision support for traffic management and emergency response.

[0084] Specifically, if Figure 2 , 3 As shown, the image definition enhancement module includes an image input submodule, a resolution processing submodule, and a refined image output submodule;

[0085] The image input submodule loads the image data based on the initial resolution road image, uses the PIL library to read the image, resizes the image to 256x256 pixels, adjusts the format to meet the input requirements of the deep learning model, and generates the processed road image;

[0086] The resolution processing submodule uses a deep residual network to enhance the resolution based on the processed road image. The network is built using Python TensorFlow. The residual block is set to include multiple 3x3 convolutional layers. The activation function uses a linear rectification function. By adding multiple residual blocks, the image resolution is gradually improved to generate an enhanced resolution reconstructed image.

[0087] The refined image output submodule reconstructs the image based on the enhanced resolution, uses the Gaussian blur algorithm of the open source computer vision library for smoothing, reduces image noise, retains key texture information, applies the Canny edge detection algorithm to highlight the information of the damaged area, and enhances the visibility of the damage by identifying the boundaries of the damaged area in the image to generate a refined road image.

[0088] In the image input submodule, the system first needs to load the road image of the initial resolution for preprocessing to meet the input requirements of the deep learning model. The Python programming language and PIL (Python Imaging Library) library are used. The process includes reading the image file and resizing it to a uniform size of 256x256 pixels. The image size is adjusted using the resize method in the PIL library, which accepts an image object and a target size parameter (in this case (256, 256)) and applies a bilinear interpolation algorithm to maintain image quality, ensuring that all input images have uniform dimensions and providing a standardized input format for subsequent processing steps. Through this preprocessing, the generated road image meets the requirements of the deep learning model in both size and format, laying the foundation for efficient image analysis.

[0089] In the resolution processing submodule, a deep residual network (ResNet) is used for resolution enhancement. The network is built using the TensorFlow library. The network gradually improves the image resolution by adding multiple residual blocks. Each residual block includes multiple 3x3 convolutional layers, which are used to extract image features, and ReLU (linear rectification function) is used as the activation function after each layer to introduce nonlinear processing and enhance the network's expressiveness. The design of the residual block allows the model to learn additional feature enhancement information instead of learning from scratch, which helps avoid the gradient vanishing problem during training. By gradually increasing the residual blocks, the network can improve the resolution while maintaining the original image content. The generated enhanced resolution reconstructed image is richer in details, providing higher quality input for subsequent image analysis.

[0090] In the refined image output submodule, the image is reconstructed based on the enhanced resolution, and the image is further processed by applying the Gaussian blur algorithm and the Canny edge detection algorithm. Gaussian blur is implemented through the computer vision library. The algorithm applies a Gaussian filter to the image to smooth the image and reduce noise. The size and standard deviation parameters of the filter are carefully selected to ensure that key texture information is retained. The Canny edge detection algorithm is used to identify the boundaries of damaged areas in the image. The Canny algorithm first applies a series of image processing steps, including Gaussian blur to eliminate noise, and then uses the gradient operator to find the intensity and direction of the edge in the image. Finally, the true edge is determined through non-maximum suppression and dual threshold detection. The refined road image retains key texture information while enhancing the visibility of the damaged area, providing high-quality image resources for road damage detection and assessment.

[0091] A specific embodiment relates to a highway detection data analysis system, in which the initial image data collected by the system is stored in RGB format, each image file size is about 3MB, and the resolution is 1024x1024 pixels. After being processed by the image input submodule, the image is adjusted to 256x256 pixels, providing input for the resolution processing submodule. The deep residual network model used by the resolution processing submodule has 50 residual blocks, each of which includes two 3x3 convolutional layers, and a batch normalization layer followed by a ReLU activation function. The resolution of the image processed by resolution enhancement is increased to 512x512 pixels, providing a basis for the refined image output submodule. In the refined image output submodule, Gaussian blur (filter size is 5x5, standard deviation is 1.5) is first applied, and then the Canny edge detection algorithm (low threshold 50, high threshold 150) is used to process the image. The refined road image finally generated highlights the cracks and potholes on the road surface. These images provide detailed visual information for road condition assessment, which helps to identify and analyze road damage, thereby improving the efficiency of road maintenance and management.

[0092] Specifically, if Figure 2 , 4 As shown, the spectrum analysis module includes a spectrum image acquisition submodule, a spectrum feature analysis submodule, and a material condition identification submodule;

[0093] The spectral image acquisition submodule collects spectral data based on the refined road image, uses a hyperspectral camera to capture the spectral characteristics of the road surface in the wavelength range of 400-1000nm, obtains the reflection and absorption spectral information of the road surface material, analyzes the road conditions, and generates spectral data images;

[0094] The spectral feature analysis submodule extracts spectral features based on the spectral data image and uses principal component analysis to extract key features of the spectral data. The principal component analysis converter sets the number of principal components to 10, retains key variation information in the spectral data, and generates spectral feature information.

[0095] The material condition identification submodule identifies the material condition based on the spectral feature information. It uses a support vector machine to build a support vector machine model in the scikit-learn library, sets the parameter C to 1.0, selects RBF as the kernel function, identifies road wear and oil stains, and generates material condition analysis results.

[0096] In the spectral image acquisition submodule, the system uses a hyperspectral camera to collect spectral data from the refined road image. The spectral camera is configured to capture spectral features within the wavelength range of 400-1000nm to obtain spectral information of the road surface material. The camera first captures a continuous spectral image of the road surface according to the set wavelength range. During the capture process, the camera's internal sensor distinguishes light of different wavelengths and records the spectral intensity of reflection and absorption. These data are stored in the order of spectral wavelengths to form a multidimensional data set. Each pixel contains complete spectral information from 400nm to 1000nm, forming a detailed spectral data image. This data image reflects the spectral characteristics of the road surface material in detail, such as spectral reflectance and absorptivity at specific wavelengths. The spectral data provides a basis for subsequent material analysis and condition assessment, including road wear, cracks, and oil pollution information. The spectral data image generated by this submodule is used to deeply analyze the physical and chemical state of the road surface, which has important guiding significance for road maintenance and repair.

[0097] In the spectral feature analysis submodule, the system extracts features from the spectral data image using the principal component analysis (PCA) technique. PCA is a statistical method used to simplify the complexity of the data set while retaining the most critical information. The PCA algorithm processes the spectral data image to reduce the dimension of the data and highlight important features. The process first calculates the covariance matrix of the data and then finds the eigenvalues ​​and eigenvectors of the matrix. These eigenvectors represent the main variation directions in the data, and the size of the eigenvalues ​​indicates the importance of each direction. When setting up the PCA converter, the number of principal components is set to 10, which means that the top 10 most important eigenvectors are selected to represent the data. These principal components can effectively retain the key variation information in the original spectral data while removing unnecessary noise and redundant information. After PCA processing, the generated spectral feature information is more focused on the key spectral features of road materials, providing a reliable data basis for the accurate identification of material conditions, significantly improving the availability and analysis efficiency of spectral data, and laying a solid foundation for the next step of material condition identification.

[0098] In the material condition recognition sub-module, a Support Vector Machine (SVM) model is constructed using the scikit-learn library to recognize the road material condition based on spectral feature information. SVM is an effective classification tool that can handle high-dimensional data and find the optimal boundary between data classes. First, the spectral feature data processed by PCA is used as input. The SVM model maps these data to a high-dimensional space through the kernel method and searches for the optimal separation hyperplane in this space. During the model construction process, the kernel function is selected as the Radial Basis Function (RBF) kernel, and the parameter C (penalty coefficient) is set to 1.0. The RBF kernel can handle non-linear data relationships, while the parameter C controls the model's tolerance for classification errors. Through the training process, the SVM model learns how to classify the road material condition based on spectral features and identify different types of road damages, such as cracks and oil stains. After training, the model can effectively classify new spectral feature data and generate material condition analysis results. These results not only provide a detailed description of the road material condition but also guide subsequent road maintenance and repair work. Through accurate material condition recognition, this sub-module greatly improves the efficiency and quality of road maintenance.

[0099] Suppose the highway detection data analysis system obtains a set of spectral image data containing multiple samples under different road conditions. The spectral data format of each sample is a multi-dimensional array in the range of 400 - 1000 nm, where each dimension represents the spectral intensity at different wavelengths. For example, a sample contains data with a reflectance of 0.5 at a wavelength of 450 nm and a reflectance of 0.7 at a wavelength of 550 nm. In the spectral feature analysis sub-module, PCA processes these data and extracts 10 principal components representing the main variations. Suppose the eigenvector corresponding to one of the principal components is [0.2, -0.3, 0.4,...], which reflects the main change trend of the road material spectral features in this dimension. In the material condition recognition sub-module, the SVM model is trained based on these features. Suppose the class of a recognized road damage sample is "oil stain", which indicates that the model has successfully classified the sample based on spectral features. Through these operations, the highway detection data analysis system can accurately identify and analyze various conditions on the road surface and provide valuable information to guide road maintenance and repair work.

[0100] Specifically, as Figure 2 、 5 shown, the enhanced real-time assistance module includes a damage data integration sub-module, an AR view generation sub-module, and a maintenance assistance display sub-module;

[0101] Based on the results of the material condition analysis, the damage data integration sub-module collects and integrates road damage data, uses DataFrame in the Pandas library to organize the data, including screening key damage features such as type, location, and degree, applies data standardization processing to unify the data format, and generates comprehensive road damage data;

[0102] Based on the comprehensive road damage data, the AR view generation sub-module creates an augmented reality view, uses Unity 3D combined with the Vuforia software development kit to build an AR scene, dynamically binds real-time road damage data to virtual objects through a scripting language, adjusts the size, position, and color of the virtual objects to match the real-time damage situation, and synchronously updates the three-dimensional interactive view to generate a real-time AR maintenance view;

[0103] Based on the real-time AR maintenance view, the maintenance assistance display sub-module conducts auxiliary display of the maintenance plan, uses OpenCV for image processing, identifies the edges and shapes of the damaged areas, combines with the virtual markers in the AR view, provides damage location and maintenance direction guidance, and generates an enhanced real-time maintenance assistance view.

[0104] In the damage data integration sub-module, the system uses DataFrame in the Pandas library to collect and integrate the results of the material condition analysis. This module receives a dataset containing road damage features, and the dataset includes key information such as damage type, location, and degree. For example, a data entry contains a damage type of "crack", location coordinates of "X: 102, Y: 438", and a degree of "medium". This data is stored in a structured format in the DataFrame for convenient subsequent processing. The module performs data screening and standardization processing. Data screening focuses on identifying and retaining the information that is most critical for road maintenance, such as paying attention to damage of a greater degree. Data standardization ensures that data from different sources and formats are converted into a unified format. For example, all location coordinates are converted into a unified reference system, and the descriptions of the damage degree are unified into levels such as "slight", "medium", and "severe". These operations are implemented through the data processing functions of Pandas. This module generates a comprehensive and standardized integrated road damage data, which will provide accurate and reliable data support for subsequent AR view generation and maintenance assistance display.

[0105] In the AR view generation sub-module, the system uses Unity 3D and the Vuforia software development kit to create an augmented reality view. This sub-module first constructs a basic 3D scene in Unity 3D as the background of the augmented reality view. Through the API of the Vuforia software development kit, real-time road damage data is dynamically bound to virtual objects. Each damage data entry is converted into a virtual object. For example, a crack damage can be represented as a 3D model of a crack pattern. Then, a scripting language (such as C#) is used to control the properties of these virtual objects, including size, position, and color, to ensure that the actual road damage situation is accurately reflected. For example, the script adjusts the length and width of the virtual crack model according to the damage degree, and places the model in the AR scene according to the position data. This sub-module updates the 3D interactive view in real time to ensure that the AR view is updated accordingly as the on-site situation changes. The generated real-time AR maintenance view provides an intuitive and interactive display of road damage for maintenance personnel, greatly improving the efficiency and accuracy of road maintenance.

[0106] In the maintenance assistance display sub-module, based on the real-time AR maintenance view, OpenCV is used for image processing to assist in the display of maintenance plans. First, image recognition is performed on the road damage area in the AR view, mainly focusing on the edges and shapes of the damage. The OpenCV library provides a series of image processing functions, such as edge detection, image segmentation, and shape recognition, which can effectively identify and analyze the characteristics of the damage area. For example, the Canny edge detection algorithm is used to identify the contour of the damage area, and then contour analysis is used to determine the shape and size of the damage. This information is used to enhance the AR view by adding virtual markers or indicating arrows around the damage area to highlight the damage location. This module also provides corresponding maintenance direction guidance according to the type and degree of the damage, combined with the information in the maintenance database. For example, for a damage identified as "severe crack", the corresponding maintenance methods and required materials will be displayed in the AR view. The generated enhanced real-time maintenance assistance view not only provides accurate positioning of the damage, but also provides practical maintenance plans, providing strong support for the decision-making of maintenance personnel.

[0107] The detailed embodiments of the above sub-modules are as follows. Suppose the damage data integration sub-module receives a series of damage data, including different types of damage such as cracks and potholes. After processing by Pandas, a standardized DataFrame containing damage type, location, and degree is obtained. For example, a piece of data is {"type": "crack", "location": "X: 102, Y: 438", "degree": "medium"}. In the AR view generation sub-module, this data corresponds to generating a 3D model of a medium-sized crack located at X: 102, Y: 438. The maintenance assistance display sub-module further uses OpenCV to identify the specific shape and size of this crack and adds auxiliary marks in the AR view. When the maintenance personnel view the real-time AR maintenance view, they can intuitively understand the specific situation of each damage and the recommended maintenance methods.

[0108] Specifically, as Figure 2 、 6 shown, the sign recognition and evaluation module includes an image recognition sub-module, a status evaluation sub-module, and an evaluation result output sub-module;

[0109] The image recognition sub-module performs image recognition of road signs based on the enhanced real-time maintenance assistance view. It uses the Python TensorFlow library to build a convolutional neural network model, defines the model structure including multiple Conv2D layers and MaxPooling2D layers to extract image features, sets the optimizer as Adam, the loss function as categorical_crossentropy, conducts model training and verification, and generates road sign recognition information;

[0110] The status evaluation sub-module evaluates the wear and occlusion status of the sign based on the road sign recognition information. It uses image processing algorithms, applies the Sobel operator for edge detection, and performs HSV color space conversion to analyze the color saturation and brightness of the sign, evaluates the visibility and integrity of the sign, and generates the road sign integrity analysis result;

[0111] The evaluation result output sub-module outputs the final evaluation result based on the road sign integrity analysis result. It uses the Python Matplotlib library for data visualization, integrates evaluation metrics, forms an evaluation chart, provides a comprehensive description of the sign status, and generates the sign status evaluation result.

[0112] In the image recognition sub-module, the system uses the Python TensorFlow library to build a Convolutional Neural Network (CNN) model to identify road signs in the enhanced real-time maintenance assistance view. The data input format of this model is preprocessed image data, where each image data represents a visual sample of a road sign. The definition of the model structure includes multiple convolutional layers (Conv2D) and max pooling layers (MaxPooling2D). The Conv2D layers are responsible for extracting features from the images, such as edges and textures. After each Conv2D layer, a MaxPooling2D layer follows, which is used to reduce the spatial dimension of the features and improve the computational efficiency of the model. When building the model, the input layer is first initialized to match the size and color channels of the preprocessed images, and then several Conv2D layers and MaxPooling2D layers are added. The parameters of these layers, such as the number and size of the filters, need to be adjusted according to the specific application scenario. In the last part of the model, a fully connected layer (Dense) is added for classification. The optimizer is selected as Adam, which is an efficient optimization algorithm with low computational overhead and is suitable for training with large amounts of data. The loss function is set to categorical_crossentropy, which is suitable for multi-classification problems. During the model training process, the fit method is used for iterative training, and at the same time, the model performance is verified through a series of validation data sets. This sub-module generates road sign recognition information, including sign types and confidence levels, which are of great significance for subsequent status evaluation and traffic management.

[0113] In the status evaluation sub-module, based on the road sign recognition information, image processing algorithms are used to evaluate the wear and occlusion status of the signs. The Sobel operator is applied for edge detection to determine the contour and shape changes of the signs, which helps to evaluate the physical wear degree of the signs. The Sobel operator highlights the edges by calculating the spatial gradient of the image grayscale and can effectively detect the clarity and integrity of the sign boundaries. The image is converted to the HSV color space to analyze the saturation and lightness of the sign colors, which is very effective for evaluating the fading and covering of the sign colors. The HSV color space conversion makes the analysis of color characteristics more intuitive and accurate and is more suitable for dealing with problems of color saturation and lightness compared to the RGB color space. Through these processing steps, this sub-module can comprehensively evaluate the visibility and integrity of road signs and generate road sign integrity analysis results. These results are crucial for road safety and maintenance and can guide road maintenance personnel to replace or repair damaged signs in a timely manner.

[0114] In the evaluation result output sub-module, based on the road sign integrity analysis results, the final evaluation results are output and displayed. This sub-module uses the Python Matplotlib library for data visualization, integrates the evaluation metrics, and forms an intuitive evaluation chart. First, it collects the integrity analysis data generated by the status evaluation sub-module. This data includes the wear degree of the signs, the changes in color saturation and lightness. Based on this data, various charts are created, such as bar charts, line charts, or pie charts, to show the distribution of different types and statuses of road signs. During the process of generating the charts, the Matplotlib library provides rich customization options, such as title, labels, color, and style settings, to ensure that the charts are clear, beautiful, and information-rich. The sign status evaluation results generated by this sub-module are presented in the form of charts, providing an intuitive decision-making support tool for road maintenance management, which helps to improve road safety and maintenance efficiency.

[0115] Suppose the image recognition sub-module receives a series of road sign images in the enhanced real-time maintenance assistance views. One of the images shows a stop sign. During the model training and validation process, this image is preprocessed and then input into the CNN model. The model successfully recognizes the sign as a "stop sign" with a confidence level of 0.95. In the status evaluation sub-module, through the Sobel operator and HSV color space conversion analysis, it is found that the edge of the sign is clear but the color is slightly faded. In the evaluation result output sub-module, the data is visualized as a chart, showing that the wear degree of this stop sign is "slight" and the color saturation has decreased by 20%, providing clear maintenance guidelines for road maintenance personnel.

[0116] Specifically, as Figure 2 、 7 shown, the sound recognition monitoring module includes a sound data acquisition sub-module, a sound event recognition sub-module, and an abnormal event result sub-module;

[0117] Based on the sign status evaluation results, the sound data acquisition sub-module collects sound data. Through sound signal processing technology, in the field of digital signal processing, the Butterworth filter is used to remove background noise, and the Fourier transform is used to extract the spectral features of the sound signal, so that the data quality meets the requirements of subsequent processing and generates the processed sound data;

[0118] Based on the processed sound data, the sound event recognition sub-module performs the recognition of sound events. It applies deep learning algorithms to build a deep neural network through Python and the TensorFlow library. The network structure is set to include multiple convolutional layers Conv2D for extracting the time-frequency features of the sound signal, processes the temporal information of the sound through the recurrent layer, and trains to recognize the target sound patterns of vehicle honking and emergency vehicle alarms, generating the sound event recognition results;

[0119] Based on the results of sound event recognition, the abnormal event result sub-module uses a data aggregation method and Python to integrate and analyze the recognized sound events, identify and evaluate traffic anomalies, including sudden accidents and emergency braking sounds, judge the abnormal changes in traffic flow, and generate traffic anomaly monitoring results.

[0120] In the sound data acquisition sub-module, through advanced digital signal processing technology, the system first preprocesses the sound data collected from the environment to ensure that the data quality meets the requirements of the subsequent deep learning model. The sound data is stored in the waveform file format (WAV), retaining the original sound signal quality and facilitating processing. The collected sound data is passed through a Butterworth filter to remove background noise. This filter is selected for its flat frequency response, effectively retaining the main features of the sound signal while removing irrelevant background noise. During the filtering process, an appropriate cut-off frequency is selected to ensure that only signals within the target frequency range are retained, thus optimizing the clarity and usability of the signal. The fast Fourier transform (FFT) is used to perform spectral analysis on the filtered sound signal, extracting the frequency-domain features of the sound signal. The FFT conversion transforms the sound signal from the time domain to the frequency domain, making it easier to identify and analyze the characteristic frequencies of sound events. Through this series of refined processing steps, clear, denoised sound data with rich spectral information is generated, laying the foundation for the accurate recognition of sound events.

[0121] In the sound event recognition sub-module, the preprocessed sound data is analyzed and recognized through deep learning technology. A deep neural network constructed using the Python programming language and the TensorFlow library, specifically the convolutional neural network (CNN) structure, is used to extract the time-frequency features in the sound signal. The network structure design includes multiple convolutional layers (Conv2D), which can capture complex patterns and features in the sound data, such as the specific frequencies and waveforms of vehicle horns and emergency vehicle alarms. Recurrent layers (such as LSTM) are also integrated into the network to process the temporal information of the sound signal, enhancing the model's ability to recognize the temporal features of sound events. During the model training process, a large amount of labeled sound data is used for supervised learning. The optimizer is selected as Adam, and the loss function uses categorical_crossentropy to adapt to the multi-classification problem of sound event recognition. Through training, the model can accurately identify and distinguish different sound events, generating sound event recognition results, which are crucial for the traffic anomaly monitoring system as they provide timely and accurate information to identify potential traffic safety issues.

[0122] In the abnormal event result sub-module, the recognition results of sound events are further processed and analyzed through data aggregation and analysis methods. Python is used for data processing. Combining the recognized sound events, such as sudden accident sounds and emergency braking sounds, the system evaluates the abnormal changes in traffic flow. This process includes analyzing the temporal and spatial distributions of sound events to identify abnormal patterns and trends in traffic flow. Through the aggregated analysis of sound events, the system can comprehensively consider the information of multiple sound sources, thereby improving the accuracy of traffic anomaly judgment. The generated traffic anomaly monitoring results are provided in the form of reports or real-time alerts, containing detailed information about the detected abnormal events, such as event types, occurrence times, and locations. This information is very valuable for traffic management departments, providing real-time safety monitoring data, helping to quickly respond to traffic safety problems, and thus improving the safety and efficiency of road use.

[0123] The specific embodiment involves a highway detection data analysis system. The system deploys sound collection devices on the main urban traffic sections and records traffic sounds continuously for 24 hours. In a data collection cycle, the system captures and stores approximately 10GB of WAV format sound data, including various vehicle passing sounds, alarm sounds in emergencies, and background noise. The sound data collection sub-module successfully removes the background noise of non-target frequencies and extracts the key spectral features of the sound signal by applying a Butterworth filter (cut-off frequency set to 1000Hz) and FFT. The sound event recognition sub-module trains a deep neural network containing 5 convolutional layers and 2 LSTM layers, accurately identifying 95% of the target sound events, such as vehicle honking and emergency vehicle alarms. The abnormal event result sub-module conducts an aggregated analysis of the recognition results, successfully identifying several abnormal changes in traffic flow caused by emergency vehicles, providing a timely traffic anomaly monitoring report for traffic management departments. The processing and analysis process not only improves the accuracy and efficiency of traffic monitoring but also enhances the ability of traffic safety management.

[0124] Specifically, as Figure 2 , 8 shown, the weather condition impact analysis module includes an image sequence processing sub-module, a weather impact analysis sub-module, and an impact result generation sub-module;

[0125] Based on the traffic anomaly monitoring results, the image sequence processing sub-module performs image sequence processing. Using a recurrent neural network, an RNN model is constructed through the Python Keras library. The model uses a recurrent layer including LSTM to process the image sequence, analyzes the temporal change information in the sequence images, learns the impact of weather condition changes on road conditions, and generates the time-series image analysis results;

[0126] Based on the results of the sequential image analysis, the weather impact analysis sub-module conducts an analysis of the impact of weather conditions. It uses sequential analysis techniques, including autoregressive models, combines historical and real-time weather data, and performs trend analysis on historical data and real-time data response analysis through Python to predict the potential impact of weather changes on road safety and generate road safety meteorological prediction results.

[0127] Based on the road safety meteorological prediction results, the impact result generation sub-module generates impact results. It uses Python and data visualization tools such as Matplotlib to display the impact of weather changes on road conditions, integrates the analysis results, provides a display of the impact of weather changes on road safety, and generates weather condition impact results.

[0128] In the image sequence processing sub-module, a recurrent neural network (RNN) is constructed through the Keras library to process the image sequence in the traffic anomaly monitoring results. The RNN model is particularly suitable for processing time series data because it can capture the temporal dynamics in the sequence. The input format of the model is a series of images sorted in time, and each image captures the road conditions at a certain time point. The core of the model is the long short-term memory (LSTM) layer, which is a special type of RNN layer that can effectively learn long-term dependencies. When building the model, one or more LSTM layers are first initialized, and the parameters of each layer, such as the number of units and activation functions, need to be adjusted according to the specific task. The LSTM layer reads the image sequence and learns the temporal change information in the sequence, such as how weather conditions that change over time affect road conditions. After training, the model can predict future road conditions based on the image sequence and generate sequential image analysis results, which are of great significance for understanding the changes in road states under different weather conditions and provide key insights for road maintenance and safety management.

[0129] In the weather impact analysis sub-module, through sequential analysis techniques, based on the sequential image analysis results and historical and real-time weather data, it conducts an analysis of the impact of weather conditions. Sequential analysis techniques, especially autoregressive models, are suitable for predicting trends and patterns in time series data. First, historical weather data and real-time weather data are collected, and these data formats include temperature, humidity, and rainfall. The autoregressive model is used to analyze the relationship between these data and the image sequence analysis results. For example, the model may find that when rainfall increases, the road conditions captured in the road image sequence become more slippery. Through trend analysis of historical data and real-time data response analysis, this sub-module can predict the potential impact of weather changes on road safety. This sub-module generates road safety meteorological prediction results, which are crucial for timely adjusting road maintenance strategies and traffic management measures.

[0130] Based on the road safety meteorological prediction results, the impact result generation sub-module uses Python and data visualization tools to display the impact of weather changes on road conditions. The Matplotlib library is used to create various charts, such as line charts, bar charts, or scatter plots, to intuitively show how weather conditions affect road safety. For example, a line chart shows the relationship between rainfall and the incidence of road accidents. By integrating the analysis results and charts, this sub-module provides a comprehensive display of the impact of weather changes on road safety, assisting road management departments in making more well-founded decisions. The generated impact results of weather conditions not only reveal how the weather affects road conditions but also predict potential risks of future weather changes, playing an important role in enhancing road safety and reducing accidents.

[0131] Suppose the image sequence processing sub-module receives a sequence of road images at different time points each day within a week, where each image records the degree of road slipperiness. After analyzing these image sequences through the LSTM model, the model learns that during continuous multi-day rainfall, the degree of road slipperiness gradually increases. In the weather impact analysis sub-module, combining the rainfall data within a week, the autoregressive model predicts a positive correlation between the increase in rainfall and the increase in road slipperiness. The impact result generation sub-module creates a line chart through Matplotlib to intuitively show the relationship between rainfall and road slipperiness within a week, providing valuable data support for road maintenance and traffic management.

[0132] Specifically, as Figure 2 , 9 shown, the road environment mapping module includes an image segmentation sub-module, an environment recognition sub-module, and a map update sub-module;

[0133] Based on the impact results of weather conditions, the image segmentation sub-module performs image segmentation. It uses a fully convolutional network and constructs the network using the Python TensorFlow library. The network includes convolutional layers for extracting image features and upsampling layers for increasing image resolution, classifying the image at the pixel level to distinguish environmental elements such as roads, vehicles, and pedestrians, and generating image segmentation results;

[0134] Based on the image segmentation results, the environment recognition sub-module uses the U-Net model for environment recognition. It constructs the U-Net model using the Keras library, improves the accuracy of image segmentation by splicing feature maps, distinguishes and identifies different elements in the road environment, including roads, vehicles, and pedestrians, and generates environment recognition results;

[0135] Based on the environment recognition results, the map update sub-module uses geographic information system technology to integrate the identified environmental elements, including roads, vehicles, and pedestrians, into the map data, update and optimize the spatial data, reflect the real-time road environment conditions, and generate a road environment map.

[0136] In the image segmentation sub-module, through advanced Fully Convolutional Network (FCN) technology, the system performs precise pixel-level classification on road images affected by weather conditions. The network built using the Python programming language and the TensorFlow library takes high-resolution RGB images as input, which are directly obtained from surveillance cameras or weather condition monitoring systems. The design of the FCN model includes multiple convolutional layers for extracting multi-level features of the image. These convolutional layers effectively analyze complex road scenes by learning local features in the image, such as edges, textures, and shapes. Through the upsampling layer, these feature maps are spatially expanded to restore the resolution of the original image, ensuring that each pixel can be accurately classified. The FCN is particularly suitable for pixel-level image understanding tasks because it directly outputs a classification result with the same size as the input image, achieving precise differentiation of environmental elements such as roads, vehicles, and pedestrians. The generated image segmentation results provide important basic data for subsequent environmental recognition. Through this high-precision image segmentation, the system can better understand and analyze the road environment, providing strong technical support for traffic monitoring and management.

[0137] In the environmental recognition sub-module, by applying the U-Net model, the system further improves the accuracy of image segmentation and achieves in-depth recognition of differentiated elements in the road environment. The U-Net model is built using the Keras library, and its unique feature is the "U" - shaped design of the model structure, which includes a contracting path and a symmetric expanding path. This design enables the model to accurately locate objects in the image while maintaining the image context information. The concatenation operation of feature maps is another core feature of the U-Net, allowing direct transfer from the contracting path to the corresponding expanding path of the model, retaining more location information, thus greatly improving the segmentation accuracy. In this way, the U-Net can effectively distinguish and identify various elements in the road environment, including different types of vehicles, pedestrians, and different states of the road itself. The generated environmental recognition results not only have a high accuracy rate but also are richer in details, providing a solid foundation for the comprehensive analysis of the road environment.

[0138] In the map update sub-module, the system adopts Geographic Information System (GIS) technology to integrate the refined environmental element information obtained from the environmental recognition sub-module, such as roads, vehicles, pedestrians, etc., into the map data. This process involves the processing and updating of spatial data, aiming to reflect the real-time captured road environment status on the map to provide the latest road usage and environmental conditions. The application of GIS technology includes multiple aspects such as data integration, spatial analysis, and map production, and can process a large amount of spatial location information and related attribute data. Through precise spatial data update and optimization, the generated road environment map not only shows the current road conditions but also contains important information such as traffic flow, vehicle type distribution, and pedestrian density. These information are of great significance for multiple aspects of traffic planning, road maintenance, and emergency response, enabling traffic management departments to make more effective decisions based on the latest road environment information.

[0139] The specific embodiment involves an urban traffic monitoring system that integrates weather monitoring data and real-time traffic images and executes the above-mentioned processing processes of image segmentation, environmental recognition, and map update through a high-performance computing platform. For example, during a heavy rainfall event, the system successfully identified waterlogging areas, moving vehicles, and pedestrians taking shelter from the rain by analyzing road images affected by the weather and updated this information in real time to the urban traffic map. This not only provides immediate road safety tips for drivers but also assists urban management departments to respond quickly and deploy relevant safety measures, effectively reducing the traffic impact under adverse weather conditions.

[0140] Specifically, as Figure 2 、 10 shown, the dynamic monitoring and anomaly detection module includes a dynamic data acquisition sub-module, a real-time monitoring and analysis sub-module, and an anomaly recognition and notification sub-module;

[0141] Based on the road environment map, the dynamic data acquisition sub-module collects dynamic road data, applies the optical flow estimation method, uses the Pyramid Lucas-Kanade function in the Python computer vision library to analyze the movement of pixels between consecutive image frames, evaluates the movement direction and speed of objects, and generates dynamic road data;

[0142] Based on the dynamic road data, the real-time monitoring and analysis sub-module uses the Isolation Forest algorithm for anomaly behavior detection, sets the parameters of the Isolation Forest model, including the number of trees and the number of samples, uses the Isolation Forest function in the Python scientific computing library to identify the anomaly points in the data, and generates the anomaly behavior recognition result;

[0143] Based on the results of abnormal behavior recognition, the abnormal recognition and notification sub-module integrates the data of abnormal behaviors, uses data aggregation technology to summarize the abnormal information, forms real-time monitoring results, sends alarm notifications through the communication interface to remind relevant personnel to carry out emergency responses, and generates real-time monitoring and abnormal notification results.

[0144] In the dynamic data collection sub-module, by applying the optical flow estimation method, especially the Pyramid Lucas-Kanade function in the Python computer vision library, the dynamic road data collection based on the road environment map is performed. Optical flow estimation is a method of analyzing the pixel movement between consecutive image frames to evaluate the movement direction and speed of objects. The input data format is a series of images sorted by time, and each image captures an instant of the road environment. These images are processed using the Pyramid Lucas-Kanade function, which calculates the optical flow vectors by tracking the movement of specific feature points at different resolution levels of the image pyramid. These vectors depict the movement paths of the feature points from one image frame to another, thereby estimating the movement direction and speed of the objects. By analyzing these optical flow vectors, this module can generate dynamic road data, including the movement patterns of vehicles and pedestrians, which provides important information for further traffic flow analysis and traffic safety management.

[0145] In the real-time monitoring and analysis sub-module, the Isolation Forest algorithm is used to detect abnormal behaviors based on the dynamic road data. The Isolation Forest is an efficient anomaly detection algorithm, especially suitable for processing high-dimensional data. In this module, first, the key parameters of the Isolation Forest model are set, including the number of trees and the number of samples, which directly affect the performance and accuracy of the model. The dynamic road data is processed using the Isolation Forest function in the Python scientific computing library. By randomly selecting features and split values to "isolate" the sample points, the abnormal points are more likely to be isolated due to their rarity. By analyzing the dynamic road data, this model can effectively identify the samples with abnormal behaviors, such as vehicles moving abnormally fast or pedestrians walking on abnormal paths. The generated abnormal behavior recognition results are crucial for timely discovering potential traffic safety problems.

[0146] Based on the results of abnormal behavior recognition, the abnormal recognition and notification sub-module performs the integration of abnormal behavior data and the generation of real-time monitoring results, and uses data aggregation technology to summarize various detected abnormal behavior information. For example, the abnormal behaviors occurring multiple times at the same location are integrated into a comprehensive message to provide a more comprehensive monitoring view. The real-time monitoring results and alarm notifications are sent to relevant personnel, such as the traffic management department and the emergency response team, through the communication interface. This process includes not only the detailed information of the abnormal behaviors but also the recommended response measures, enabling relevant personnel to respond quickly and take necessary safety and management measures, thereby effectively improving road safety and emergency response capabilities.

[0147] Suppose the dynamic data acquisition sub-module collects a continuous image sequence of a busy intersection, analyzes these images through the Pyramid Lucas-Kanade function, and identifies a vehicle moving abnormally fast. In the real-time monitoring and analysis sub-module, the Isolation Forest algorithm further analyzes this dynamic data and successfully identifies the behavior of this vehicle as abnormal. Then, in the anomaly identification and notification sub-module, this anomaly behavior information is integrated and sent to the traffic management center through the communication interface to warn of the existing safety risks. Through these steps, the highway detection data analysis system can not only monitor the road conditions in real time, but also timely warn the relevant departments about potential safety threats, thereby improving the overall safety of road use.

[0148] Although the present application has been described in conjunction with specific features and their embodiments, it is obvious that various modifications and combinations can be made without departing from the spirit and scope of the present application. Accordingly, this specification and the drawings are merely exemplary illustrations of the present application as defined by the appended claims, and are considered to have covered any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. Highway detection data analysis system, characterized in that: The system includes an image clarity enhancement module, a spectrum analysis module, an enhanced real-time assistance module, a sign recognition and evaluation module, a sound recognition monitoring module, a weather condition impact analysis module, a road environment mapping module, and a dynamic monitoring and anomaly detection module; The image definition enhancement module is based on the initial resolution road image and adopts a deep residual network resolution algorithm. It builds a deep residual block to learn the mapping relationship between resolution images, improves the image reconstruction quality by increasing the depth of the network, outputs a resolution image, identifies and analyzes the damage on the road surface, and generates a refined road image. The spectral analysis module is based on the refined road image, adopts the hyperspectral imaging technology and the support vector machine algorithm, collects the multi-dimensional spectral data of the road surface to capture the characteristics of the reflected and absorbed light of the differentiated materials, uses the SVM to classify the multi-dimensional data, identifies the spectral signatures of the differentiated materials, monitors the road wear and oil pollution, and generates the material condition analysis results; The enhanced real-time assistance module uses augmented reality technology based on the material condition analysis results, combines real-time road damage data with the user's field of view, and uses mobile devices and AR glasses to overlay and display virtual information, assisting maintenance personnel in locating the damage location and providing maintenance solutions, thereby improving maintenance efficiency and accuracy and generating an enhanced real-time maintenance assistance view; The sign recognition and assessment module is based on the enhanced real-time maintenance auxiliary view and adopts a convolutional neural network model to identify road signs from images through deep learning technology, and assess their wear and occlusion status, thereby improving the accuracy of sign recognition, assessing the visibility and integrity of the signs, and generating a sign status assessment result; The sound recognition monitoring module uses sound signal processing technology and deep neural network based on the sign status evaluation results to identify the target sound pattern by performing frequency analysis on the sound signal and training the deep learning model, thereby improving the detection rate of traffic anomaly events and generating traffic anomaly monitoring results; The weather condition impact analysis module is based on the traffic anomaly monitoring results and adopts a recurrent neural network model to learn the impact of weather condition changes on road conditions by analyzing the time information in the sequence images, and comprehensively processes and analyzes historical and real-time weather data through a time series analysis method to predict the potential impact of weather changes on road safety and generate weather condition impact results; The road environment mapping module uses a full convolutional network and U-Net model to classify images at the pixel level based on the results of weather conditions, distinguish between roads, vehicles, and pedestrians, and improve the accuracy of environmental recognition through image segmentation technology to generate a road environment map; The dynamic monitoring and anomaly detection module is based on the road environment map, adopts the optical flow estimation method and the isolation forest algorithm, evaluates the movement direction and speed of the object by analyzing the pixel movement between consecutive image frames, and uses the isolation forest algorithm to detect abnormal behavior. It combines the dynamic monitoring and anomaly detection methods to monitor road usage in real time and identify abnormal behavior, and generate real-time monitoring and anomaly notification results.

2. The highway detection data analysis system according to claim 1, characterized in that: The refined road image includes road surface damage features, road markings, and road environment details; the material condition analysis results include road wear degree, oil distribution, and spectral signatures of differentiated materials; the enhanced real-time maintenance assistance view specifically combines road damage location, maintenance plan, and virtual assistance information; the sign status assessment results include road sign recognition information, wear degree assessment, and occlusion status; the traffic anomaly monitoring results include target sound pattern recognition and abnormal event detection; the weather condition impact results include an assessment of the impact of weather changes on road conditions and a prediction of the potential impact of weather conditions on road safety in future time periods; the road environment map includes environmental elements that distinguish roads, vehicles, and pedestrians; the real-time monitoring and abnormal notification results include the direction of object movement, speed assessment, and detection of abnormal behavior.

3. The highway detection data analysis system according to claim 1, characterized in that: The image definition enhancement module includes an image input submodule, a resolution processing submodule, and a refined image output submodule; The image input submodule loads image data based on the initial resolution road image, uses the PIL library to read the image, adjusts the image size to 256x256 pixels, adjusts the format to meet the input requirements of the deep learning model, and generates a processed road image; The resolution processing submodule uses a deep residual network to perform resolution enhancement based on the processed road image, uses Python TensorFlow to build the network, sets the residual block to include multiple 3x3 convolutional layers, and uses a linear rectification function as the activation function. By adding multiple residual blocks, the image resolution is gradually improved to generate an enhanced resolution reconstructed image. The refined image output submodule reconstructs the image based on the enhanced resolution, uses the Gaussian blur algorithm of the open source computer vision library for smoothing, reduces image noise, retains key texture information, applies the Canny edge detection algorithm to highlight the information of the damaged area, and enhances the visibility of the damage by identifying the boundary of the damaged area in the image to generate a refined road image.

4. The highway detection data analysis system according to claim 1, characterized in that: The spectrum analysis module includes a spectrum image acquisition submodule, a spectrum feature analysis submodule, and a material condition identification submodule; The spectral image acquisition submodule collects spectral data based on the refined road image, uses a hyperspectral camera to capture the spectral characteristics of the road surface in the wavelength range of 400-1000nm, obtains the reflection and absorption spectral information of the road surface material, analyzes the road conditions, and generates a spectral data image; The spectral feature analysis submodule extracts spectral features based on the spectral data image, uses principal component analysis to extract key features of the spectral data, and the principal component analysis converter sets the number of principal components to 10, retains key variation information in the spectral data, and generates spectral feature information; The material condition identification submodule identifies the material condition based on the spectral feature information, uses a support vector machine, builds a support vector machine model in the scikit-learn library, sets the parameter C to 1.0, selects RBF as the kernel function, identifies road wear and oil stains, and generates material condition analysis results.

5. The highway detection data analysis system according to claim 1, characterized in that: The enhanced real-time auxiliary module includes a damage data integration submodule, an AR view generation submodule, and a maintenance auxiliary display submodule; The damage data integration submodule collects and integrates road damage data based on the material condition analysis results, and uses the DataFrame in the Pandas library to organize the data, including screening key damage features including type, location, and degree, applying data standardization processing to unify the data format, and generating comprehensive road damage data; The AR view generation submodule creates an augmented reality view based on the comprehensive road damage data, uses Unity 3D combined with the Vuforia software development kit to build an AR scene, dynamically binds the real-time road damage data to the virtual object through the scripting language, adjusts the size, position, and color of the virtual object to match the real-time damage situation, performs synchronous updates of the three-dimensional interactive view, and generates a real-time AR maintenance view; The maintenance auxiliary display submodule performs auxiliary display of maintenance plans based on the real-time AR maintenance view, uses OpenCV for image processing, identifies the edges and shapes of damaged areas, and combines virtual markers in the AR view to provide damage location and maintenance direction guidance, thereby generating an enhanced real-time maintenance auxiliary view.

6. The highway detection data analysis system according to claim 1, characterized in that: The sign recognition and evaluation module includes an image recognition submodule, a state evaluation submodule, and an evaluation result output submodule; The image recognition submodule performs image recognition of road signs based on the enhanced real-time maintenance auxiliary view, builds a convolutional neural network model using the Python TensorFlow library, defines a model structure including multiple Conv2D layers and MaxPooling2D layers to extract image features, sets the optimizer to Adam and the loss function to categorical_crossentropy, performs model training and verification, and generates road sign recognition information; The state assessment submodule assesses the wear and occlusion state of the road sign based on the road sign recognition information, uses an image processing algorithm, applies the Sobel operator to perform edge detection, and uses the HSV color space conversion to analyze the color saturation and brightness of the sign, assesses the visibility and integrity of the sign, and generates a road sign integrity analysis result; The evaluation result output submodule outputs the final evaluation result based on the road sign integrity analysis result, uses the Python Matplotlib library for data visualization, integrates the evaluation indicators, forms an evaluation chart, provides a comprehensive description of the sign status, and generates the sign status evaluation result.

7. The highway detection data analysis system according to claim 1, characterized in that: The sound recognition monitoring module includes a sound data collection submodule, a sound event recognition submodule, and an abnormal event result submodule; The sound data acquisition submodule collects sound data based on the sign status evaluation result, applies Butterworth filter to remove background noise in the field of digital signal processing through sound signal processing technology, and uses Fourier transform to extract the spectrum characteristics of the sound signal, so that the data quality meets the subsequent processing requirements, and generates processed sound data; The sound event recognition submodule performs sound event recognition based on the processed sound data, applies a deep learning algorithm to build a deep neural network through Python and TensorFlow library, sets a network structure including multiple convolutional layers Conv2D for extracting time-frequency features of sound signals, processes the timing information of the sound through a recurrent layer, and recognizes the target sound patterns of vehicle horns and emergency vehicle alarms through training, and generates sound event recognition results; The abnormal event result submodule is based on the sound event recognition results, adopts the data aggregation method, uses Python to integrate and analyze the recognized sound events, and identifies and evaluates traffic anomalies, including sudden accidents and emergency braking sounds, determines abnormal changes in traffic flow, and generates traffic anomaly monitoring results.

8. The highway detection data analysis system according to claim 1, characterized in that: The weather condition impact analysis module includes an image sequence processing submodule, a weather impact analysis submodule, and an impact result generation submodule; The image sequence processing submodule performs image sequence processing based on traffic anomaly monitoring results, adopts a recurrent neural network, and builds an RNN model through the Python Keras library. The model uses a recurrent layer including LSTM to process image sequences, analyze the time change information in the sequence images, learn the impact of weather condition changes on road conditions, and generate time series image analysis results; The weather impact analysis submodule performs weather condition impact analysis based on the time series image analysis results, uses time series analysis technology, including autoregressive models, combines historical and real-time weather data, performs historical data trend analysis and real-time data response analysis through Python, predicts the potential impact of weather changes on road safety, and generates road safety meteorological forecast results; The impact result generation submodule generates impact results based on road safety meteorological forecast results, uses Python and data visualization tools including Matplotlib to display the impact of weather changes on road conditions, integrates analysis results, provides a display of the impact of weather changes on road safety, and generates weather condition impact results.

9. The highway detection data analysis system according to claim 1, characterized in that: The road environment mapping module includes an image segmentation submodule, an environment recognition submodule, and a map update submodule; The image segmentation submodule performs image segmentation based on the results of weather conditions. It uses a fully convolutional network and uses the Python TensorFlow library to build a network. The network includes a convolution layer for extracting image features and an upsampling layer for increasing image resolution. It performs pixel-level classification on the image, distinguishes environmental elements such as roads, vehicles, and pedestrians, and generates image segmentation results. The environment recognition submodule uses the U-Net model for environment recognition based on the image segmentation results. The U-Net model is built using the Keras library, and the image segmentation accuracy is improved by splicing feature maps. Differentiated elements in the road environment, including roads, vehicles, and pedestrians, are distinguished and recognized to generate environment recognition results. Based on the environmental recognition results, the map update submodule adopts geographic information system technology to integrate the recognized environmental elements, including roads, vehicles, and pedestrians, into the map data, update and optimize the spatial data, reflect the real-time road environment conditions, and generate a road environment map.

10. The highway detection data analysis system according to claim 1, characterized in that: The dynamic monitoring and anomaly detection module includes a dynamic data acquisition submodule, a real-time monitoring and analysis submodule, and an anomaly identification and notification submodule; The dynamic data acquisition submodule collects dynamic road data based on the road environment map, applies the optical flow estimation method, uses the pyramid Lucas-Kanade function in the Python computer vision library, analyzes the movement of pixels between consecutive image frames, evaluates the movement direction and speed of the object, and generates dynamic road data; The real-time monitoring and analysis submodule uses the isolation forest algorithm to detect abnormal behavior based on dynamic road data, sets the parameters of the isolation forest model, including the number of trees and the number of samples, uses the isolation forest function in the Python scientific computing library, identifies abnormal points in the data, and generates abnormal behavior recognition results; The anomaly identification and notification submodule integrates abnormal behavior data based on the abnormal behavior identification results, uses data aggregation technology to summarize abnormal information, forms real-time monitoring results, sends alarm notifications through the communication interface, reminds related personnel to respond to emergencies, and generates real-time monitoring and abnormal notification results.

Images