A radar-aware highway monitoring data processing system and method

Abstract

The application provides a radar-aware highway monitoring data processing system and method, relates to the technical field of radar data processing, and comprises the following steps: a multi-source data acquisition module collects first radar data in a highway field and second radar data of a specified local field through satellite radar and unmanned aerial vehicle radar; a rockfall scale identification module identifies the scale characteristics of rockfall according to the second radar data, and obtains rockfall scale characteristics; a confidence analysis module analyzes the satellite radar confidence of the first radar data according to the rockfall scale characteristics and local historical radar data of the specified local field, identifies the scale characteristics of rockfall, and obtains rockfall scale characteristics; and a data processing output module analyzes the identification confidence according to the rockfall scale characteristics, the rockfall scale characteristics and field historical radar data, analyzes and compensates the rockfall disaster rate, and takes the monitoring data processing result. The technical problem of insufficient reliability of dynamic field highway disaster early warning in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of radar data processing, in particular to a radar perception highway monitoring data processing system and method. BACKGROUND

[0002] The dynamic field highway disaster early warning can effectively shorten the disaster response time and reduce the traffic accident rate by capturing sudden risks such as slope rockfall in real time.

[0003] However, in the prior art, the dynamic field highway disaster early warning relies on a single monitoring method, such as monitoring by satellite radar and ground sensors only. However, these single monitoring methods have limited monitoring range and are susceptible to weather interference, resulting in insufficient accuracy of monitoring data, which can easily lead to disaster misjudgment or omission, making the disaster early warning result unreliable and difficult to meet the safety protection needs of complex highway fields.

[0004] Therefore, there is an urgent need for a holographic perception disaster early warning scheme that can integrate multi-source data, break through the technical bottleneck of single monitoring, and improve the reliability of dynamic field highway disaster early warning. SUMMARY

[0005] The present application provides a radar perception highway monitoring data processing system and method to solve the technical problem of insufficient reliability of dynamic field highway disaster early warning in the prior art.

[0006] The technical solution of the present application to solve the above technical problem is as follows:

[0007] In a first aspect, the present application provides a radar perception highway monitoring data processing system, comprising:

[0008] A multi-source data acquisition module is configured to acquire first radar data in a highway field by satellite radar and second radar data in a specified local field in the highway field by unmanned aerial vehicle radar.

[0009] A rockfall size identification module is configured to identify the size characteristics of rockfall in the highway field based on the second radar data and obtain rockfall size characteristics.

[0010] A confidence analysis module is configured to analyze the satellite radar confidence of the first radar data based on the rockfall size characteristics and local historical radar data of the specified local field, identify the size characteristics of rockfall in the highway field, and obtain rockfall size characteristics.

[0011] A data processing output module is configured to analyze and obtain identification confidence based on the rockfall size characteristics, rockfall size characteristics, and field historical radar data, calculate rockfall confidence in combination with the satellite radar confidence, analyze and compensate rockfall disaster rate, and output as a monitoring data processing result.

[0012] Secondly, the present invention provides a radar-sensing method for processing highway monitoring data, comprising:

[0013] First radar data within the highway area is collected using satellite radar, and second radar data within a designated local area within the highway area is collected using drone radar.

[0014] Based on the second radar data, the scale characteristics of falling rocks in the highway area are identified to obtain the scale characteristics of falling rocks.

[0015] Based on the rockfall scale characteristics and local historical radar data of the specified local area, the satellite radar confidence level of the first radar data is analyzed, and the rockfall scale characteristics in the highway area are identified to obtain the rockfall scale characteristics.

[0016] Based on the characteristics of rockfall scale, rockfall size, and historical radar data of the field, the identification confidence level is obtained through analysis. The rockfall confidence level is calculated by combining the satellite radar confidence level. The rockfall disaster rate is then analyzed and compensated, which serves as the result of the monitoring data processing.

[0017] The beneficial effects of this invention are:

[0018] Compared to existing technologies, this application firstly acquires first radar data within the highway area using a multi-source data acquisition module based on satellite radar, and then acquires second radar data for a designated local area within the highway area using UAV radar. This achieves both global coverage and detailed local data acquisition, providing a reliable data foundation for subsequent rockfall feature identification, confidence analysis, and early warning decision-making. Secondly, the rockfall scale identification module identifies the scale features of rocks within the highway area based on the second radar data, obtaining rockfall scale characteristics. A deep learning model intelligently processes the second radar data for the designated local area acquired by the UAV radar, accurately extracting rockfall scale features. This solves the problems of low efficiency and large errors associated with traditional manual identification, providing reliable data support for subsequent disaster early warning decision-making. Thirdly, the confidence analysis module analyzes the satellite radar confidence of the first radar data based on the rockfall scale characteristics and local historical radar data for the designated local area, and identifies rockfall scale features, obtaining rockfall scale characteristics. The reliability of satellite data is quantified through satellite radar confidence measurement, and automated statistics of the number of rocks across the entire area are achieved, providing reliable data support for subsequent disaster early warning. Finally, the data processing output module analyzes and obtains the identification confidence level based on the rockfall scale characteristics, rockfall size characteristics, and historical radar data of the field. It calculates the rockfall confidence level by combining the satellite radar confidence level, analyzes and compensates for the rockfall disaster rate, and uses this as the result of monitoring data processing. By compensating for the rockfall disaster rate through the rockfall confidence level and combining it with the disaster rate threshold, it achieves accurate judgment of disaster early warning and improves the reliability of highway disaster early warning decision-making.

[0019] Through the aforementioned technical solution, this application achieves full coverage of highway monitoring data and detailed collection of high-risk designated local areas through collaborative perception of satellite radar and UAV radar. This enhances the comprehensiveness and relevance of data collection. Based on an artificial intelligence model, it accurately identifies the scale and size characteristics of rockfalls. Combining satellite radar confidence and identification confidence with dual-dimensional verification, it dynamically compensates for rockfall disaster rates through rockfall confidence. Finally, based on disaster rate thresholds, it achieves accurate disaster early warning. This effectively overcomes the limitations of single monitoring methods, improves the reliability and accuracy of dynamic highway disaster early warning, and can fully adapt to the safety protection needs of complex highway areas such as mountainous regions and canyons. Attached Figure Description

[0020] Figure 1 A schematic diagram of the structure of a radar-sensing highway monitoring data processing system provided by the present invention;

[0021] Figure 2 This is a flowchart illustrating a radar-sensing method for processing highway monitoring data provided by the present invention.

[0022] In the attached diagram, the components represented by each number are as follows:

[0023] Multi-source data acquisition module 11, rockfall scale identification module 12, confidence analysis module 13, and data processing output module 14. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0026] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0027] Example 1, as Figure 1 As shown, this embodiment of the invention provides a radar-sensing highway monitoring data processing system, which includes: a multi-source data acquisition module 11, a rockfall scale identification module 12, a confidence analysis module 13, and a data processing output module 14.

[0028] The multi-source data acquisition module 11 is used to acquire first radar data within the highway field via satellite radar and second radar data within a specified local area within the highway field via UAV radar.

[0029] Dynamic field highway disaster early warning can effectively shorten disaster response time and reduce traffic accident rate by capturing sudden risks such as slope rockfalls in real time. However, traditional early warning methods such as manual inspection and fixed cameras are difficult to balance large-scale coverage and local monitoring accuracy. Either the limited monitoring range leads to missed detection of disasters in remote road sections, or the insufficient resolution leads to misjudgment of small-scale disasters. The early warning has problems of missed reports, false reports, and low reliability.

[0030] To address the aforementioned issues, this application utilizes satellite radar to collect first radar data within the highway area, and uses UAV radar to collect second radar data for a designated local area within the highway area.

[0031] Specifically, the multi-source data acquisition module 11 in the system includes:

[0032] The first data acquisition unit is used to acquire first radar data within the highway area via satellite radar;

[0033] A local field determination unit is used to determine a specified local field within a highway field, wherein the specified local field is a part of the highway field;

[0034] The second data acquisition unit is used to acquire second radar data of the designated local area through the UAV radar.

[0035] In this embodiment, firstly, satellite radar collects radar data within the highway area. The highway area refers to the entire highway requiring early warning and its surrounding related areas, such as slopes and mountainsides on both sides of the highway, typically covering tens to hundreds of kilometers. Exemplarily, the satellite radar emits electromagnetic waves of a specific band into the highway area. These electromagnetic waves penetrate clouds, rain, fog, and other meteorological interference, illuminating and reflecting off targets within the highway area, such as mountainsides, road surfaces, and fallen rocks. The satellite radar receives the reflected echoes and obtains radar image data, which serves as the first radar data. The coverage of this first radar data can completely encompass the entire highway area, and the data is updated regularly, unaffected by mountain obstructions or foggy weather. It can continuously capture the distribution trends of the entire highway area. This wide coverage and strong continuity provide a macroscopic perspective for disaster early warning.

[0036] Secondly, due to the significant differences in disaster risk across different areas within the highway monitoring area, it is unnecessary to monitor the entire highway area with the same level of precision. Therefore, designated local areas are identified within the highway monitoring area. These designated local areas are part of the highway monitoring area and are selected from local regions with high disaster risk. For example, critical nodes where rockfalls are likely to cause serious consequences, such as road slopes that have historically experienced frequent rockfalls, geologically unstable mountain areas, sharp bends, or tunnel entrances / exits, can be selected as designated local areas. This concentrates high-risk monitoring resources on high-risk areas, improving monitoring efficiency and targeting.

[0037] Finally, a second radar data is collected from a designated local area using drone radar. The monitoring range of the drone radar is this designated local area, typically a small area of ​​several kilometers or even hundreds of meters, such as a 500-meter-long high-risk slope. For example, the radar echo image data generated by the drone-mounted radar equipment transmitting electromagnetic waves and receiving the reflected echoes from the designated local area, after signal processing, serves as the second radar data. Each pixel in the second radar data contains not only spatial coordinate information but also physical parameters such as echo intensity and phase. For instance, hard rocks reflect stronger electromagnetic waves and appear as high-brightness areas in the second radar data, while soft soil or vegetation corresponds to low-brightness areas. The second radar data can be presented as a two-dimensional grayscale image or a three-dimensional point cloud: a two-dimensional grayscale image visually presents the distribution of different targets through grayscale differences, while a three-dimensional point cloud accurately records the three-dimensional coordinates of each reflection point, providing a more detailed reflection of the three-dimensional outline of the rocks. Compared to the first radar data, the second radar data has higher resolution (e.g., up to centimeter level), greater flexibility (e.g., close-range slope monitoring), and can capture small-scale rockfall details that are difficult for satellite radar to identify. This provides high-precision data for subsequent identification of rockfall scale characteristics, compensating for the lack of local details caused by the resolution limitations of satellite radar.

[0038] In summary, compared to existing technologies, this application uses satellite radar to collect first radar data within the highway area, and uses UAV radar to collect second radar data within a designated local area of ​​the highway area. This achieves both global coverage and detailed local data collection, providing a reliable data foundation for subsequent rockfall feature identification, confidence analysis, and early warning decision-making.

[0039] The rockfall scale identification module 12 is used to identify the scale characteristics of rocks in the highway area based on the second radar data, and obtain the rockfall scale characteristics.

[0040] Significant differences exist in the dimensional characteristics of falling rocks, such as diameter and volume. These differences directly lead to significant variations in the degree of damage, the scope of impact, and the level of risk caused by the disasters. For example, small falling rocks with a diameter of less than 0.5 meters and a volume of less than 0.1 cubic meters may only cause minor scratches on the road surface and pose a low threat to passing vehicles. However, medium-sized falling rocks with a diameter of 1-2 meters and a volume of 1-8 cubic meters may penetrate the road structure, damage guardrails, or even force the temporary closure of a single lane. Based on this strong correlation between the dimensional characteristics of falling rocks and the consequences of the disasters, it is necessary to accurately identify the dimensional characteristics of falling rocks.

[0041] To address the aforementioned issues, this application uses the second radar data to identify the scale characteristics of falling rocks within the highway area, thereby obtaining the scale characteristics of the falling rocks.

[0042] Specifically, the rockfall scale identification module 12 in the system includes:

[0043] The scale recognition model calling unit is used to obtain the rockfall scale recognizer;

[0044] The rockfall scale identification unit is used to input the second radar data into the rockfall scale identifier and identify and output the rockfall scale features.

[0045] In this embodiment of the application, a rockfall scale identifier is first obtained. The rockfall scale identifier is constructed based on deep learning, inputs second radar data, and outputs rockfall scale features.

[0046] Secondly, the second radar data is input into the rockfall scale identifier, which identifies and outputs the rockfall scale characteristics. These characteristics refer to quantifiable physical parameters such as the diameter, volume, and surface area of ​​the rockfall, accurately representing its physical size and shape. For example, inputting the second radar data of a slope into the rockfall scale identifier identifies and outputs that a rockfall in that specified local area has a diameter of 0.75 meters.

[0047] Specifically, the scale recognition model invocation unit includes:

[0048] The data preparation unit is used to collect sample second radar data sets based on historical data of UAV radar monitoring of falling rocks, and to label all falling rock scales in each sample second radar data set to obtain sample falling rock scale feature set.

[0049] The model building unit is used to build a rockfall scale recognizer based on deep learning, wherein the input data is the second radar data and the output data is the rockfall scale features.

[0050] The model training unit is used to perform supervised training and optimization of the rockfall scale recognizer using the sample second radar data set and the sample rockfall scale feature set. After verifying that convergence is satisfied, the training is completed and the rockfall scale recognizer is obtained.

[0051] In this embodiment, firstly, based on historical data of UAV radar monitoring of falling rocks, a sample second radar data set is collected. All falling rock scales within each sample second radar data set are labeled to obtain a sample falling rock scale feature set. For example, second radar data covering various possible falling rock forms (size, shape, distribution, etc.) is collected from historical data of UAV radar monitoring of falling rocks as a sample second radar data set. Then, all falling rocks in each sample second radar data set are manually labeled, marking the actual scale characteristics of each falling rock. For example, a falling rock area is selected on a sample second radar data set and labeled with a diameter of 0.6 meters, forming a sample falling rock scale feature set. The sample second radar data set and the sample falling rock scale feature set are the training dataset for the falling rock scale recognizer.

[0052] Secondly, a rockfall scale recognizer based on deep learning is constructed, where the input data is second radar data and the output data is rockfall scale features. Specifically, radar data is essentially the reflection signal of electromagnetic waves by an object. Due to the different scales of falling rocks, the intensity and distribution of the reflected signal also differ. Therefore, these mapping rules can be learned through a deep learning model to establish a mapping relationship between the second radar data and the rockfall scale features. For example, the rockfall scale recognizer can be built based on an improved convolutional neural network (CNN) architecture. It achieves quantitative recognition of rockfall scale features through hierarchical feature extraction, multi-scale fusion, and accurate regression. The rockfall scale recognizer mainly consists of an input layer and preprocessing module, a feature extraction module, a feature fusion and regression module, and an output layer.

[0053] Optionally, preprocessing includes noise suppression (e.g., removing random noise from radar echoes through Gaussian filtering), grayscale normalization (e.g., scaling pixel values ​​to the [0,1] range), and spatial calibration (e.g., ensuring that 1 pixel corresponds to an actual physical distance of 0.02 meters through geographic coordinate mapping, establishing a direct correlation between image scale and real size). The input layer receives the preprocessed second radar data, which is a fixed-size two-dimensional image (e.g., 320×320 pixels).

[0054] Optionally, the feature extraction module consists of four progressive convolutional units, which progressively extract features strongly correlated with the rockfall scale from the second radar data. Convolutional unit 1 (shallow feature extraction) includes a 64-channel 3×3 convolutional layer (stride 1, padding 1) to capture the abrupt changes in echo intensity at the rockfall edge, as well as a batch normalization layer and LeakyReLU activation function, and a 2×2 max pooling layer (stride 2), compressing the output feature map size to 160×160×64. Convolutional unit 2 (mid-level feature extraction) includes a 128-channel 3×3 convolutional layer, focusing on the local contour features of the rockfall, such as the continuous reflection pattern of irregular edges, as well as a batch normalization layer and LeakyReLU activation function, and a 2×2 max pooling layer, compressing the output feature map size to 160×160×64. The size is 80×80×128; Convolutional Unit 3 (deep feature extraction) contains one 256-channel 5×5 convolutional layer (expanding the receptive field) to capture the overall morphological features of the falling rocks (such as the size of the area enclosed by the outline), as well as a batch normalization layer and LeakyReLU activation function, and one 2×2 max pooling layer, with an output feature map size of 40×40×256; Convolutional Unit 4 (fine feature enhancement) contains one 512-channel 3×3 convolutional layer to enhance the high echo intensity features of the core area of ​​the falling rocks (such as the dense reflection signal of hard rock), as well as a batch normalization layer and LeakyReLU activation function, and one global average pooling layer, which transforms the 40×40×512 feature map into a 512-dimensional feature vector to compress spatial information and retain global features.

[0055] Optionally, the feature fusion and regression module concatenates the 80×80×128 feature map of convolutional unit 2 with the feature vector of convolutional unit 4, compresses the dimension through a 1×1 convolutional layer with 128 channels, and obtains a 128-dimensional fused feature vector. The regression head contains two fully connected layers (with 64 and 32 hidden neurons respectively), performs non-linear mapping through the ReLU activation function, and adds a dropout layer (probability 0.25) to suppress overfitting and enhance the model's generalization ability.

[0056] Optionally, the output layer is a single-neuron linear layer that directly outputs the predicted value of the rockfall scale features.

[0057] Finally, the rockfall scale recognizer is trained and optimized in a supervised manner using the sample second radar dataset and the sample rockfall scale feature set. Training is completed after verification that convergence is achieved, thus obtaining the rockfall scale recognizer. For example, the rockfall scale recognizer can be trained using the following technical path: 1. Data preparation: The sample second radar dataset and the sample rockfall scale feature set are divided into a training set (for model parameter learning), a validation set (for hyperparameter tuning during training), and a test set (for final evaluation of the model's generalization ability) in a 7:1.5:1.5 ratio. 2. Model Training: Using the sample second radar data in the training set as input, and the corresponding sample rockfall scale features (such as rockfall diameter) as supervision labels, the root mean square error (RMSE) is selected as the loss function. The optimization objective is to minimize the difference between the predicted rockfall scale features and the sample rockfall scale features. The Adam optimizer is used, with an initial learning rate of 0.001 and a weight decay coefficient of 1e-5. When the validation set loss does not decrease for 5 consecutive rounds, the learning rate is decayed to 1 / 10 of the original, i.e., from 0.001→0.0001→0.00001, balancing the model convergence speed and parameter fine-tuning accuracy. During training, the model parameters, such as convolutional kernel weights and fully connected layer coefficients, are iteratively updated through the backpropagation algorithm. Each iteration inputs a batch of samples (batch size=32), calculates the error between the predicted rockfall scale features and the sample rockfall scale features, and adjusts the parameters through gradient descent to reduce the error until the model's fit to the training set gradually stabilizes. 3. Convergence Judgment: During the training process, the performance of the model on the validation set that did not participate in the model training needs to be continuously verified. When the root mean square error (RMSE) of the validation set is stable within the preset threshold (e.g., ≤0.05 meters) for 10 consecutive rounds, and the fluctuation range is ≤0.005 meters, the model is considered to have converged, and the rockfall scale recognizer is obtained after training.

[0058] In summary, compared to existing technologies, this application identifies the scale features of falling rocks within a highway area based on the second radar data, thereby obtaining the scale characteristics of the falling rocks. In this way, by intelligently processing the second radar data of a designated local area collected by the UAV radar using a deep learning model, the scale features of falling rocks are accurately extracted, solving the problems of low efficiency and large errors in traditional manual identification, and providing reliable data support for subsequent disaster early warning decisions.

[0059] The confidence analysis module 13 is used to analyze the satellite radar confidence of the first radar data based on the rockfall scale characteristics and the local historical radar data of the specified local area, and to identify the rockfall scale characteristics in the highway area to obtain the rockfall scale characteristics.

[0060] Due to the limitations of spatial resolution of satellite radar and its susceptibility to complex terrain and weather conditions, the reliability of first-hand radar data naturally fluctuates. Small rockfalls may cause echo ambiguity. If first-hand radar data is used directly for rockfall scale identification, it may lead to deviations in rockfall scale characteristics (such as quantity and density). For example, non-rockfall targets (such as road bumps) may be misjudged as rocks, resulting in an overestimation of the quantity, or small-scale rocks may be hidden in the background signal, resulting in an underestimation of the quantity. Ultimately, this will affect the accuracy of the overall rockfall disaster assessment.

[0061] To address the aforementioned issues, this application analyzes the satellite radar confidence level of the first radar data based on the rockfall scale characteristics and local historical radar data of a specified local area, and performs rockfall scale characteristic identification within the highway area to obtain the rockfall scale characteristics.

[0062] Specifically, the confidence analysis module 13 in the system includes:

[0063] The historical scale feature extraction unit is used to acquire radar data within a specified local field during a historical period, as local historical radar data, extract the historical rockfall scale feature set of the local historical radar data, and calculate the mean to obtain local historical rockfall scale features.

[0064] The first satellite radar confidence calculation unit is used to calculate the similarity between the rockfall scale features and the local historical rockfall scale features to obtain the first satellite radar confidence.

[0065] The fusion scale feature acquisition unit is used to calculate the mean of the rockfall scale feature and the local historical rockfall scale feature to obtain the fused local rockfall scale feature.

[0066] Minimum resolution acquisition unit, used to acquire the minimum resolution of the first radar data;

[0067] The second satellite radar confidence calculation unit is used to calculate the ratio of the fused local rockfall scale features to the minimum resolution to obtain the second satellite radar confidence, wherein the maximum second satellite radar confidence is 1.

[0068] The satellite radar confidence calculation unit is used to calculate the satellite radar confidence based on the first satellite radar confidence and the second satellite radar confidence.

[0069] In this embodiment, radar data within a specified local area over a historical period (e.g., the past 3 years) is first acquired as local historical radar data. The scale features of all falling rocks are extracted from this data as a historical falling rock scale feature set, and the mean is calculated to obtain the local historical falling rock scale features. For example, the local historical falling rock scale feature of a specified local area is calculated to be 0.8 meters in diameter. The local historical falling rock scale features can be used as a historical benchmark for the falling rock scale of the specified local area to measure the rationality of the current identification results.

[0070] Secondly, the similarity between the rockfall scale feature and the local historical rockfall scale feature is calculated to obtain the first satellite radar confidence score. The first satellite radar confidence score reflects the degree of agreement between the rockfall scale feature and the local historical rockfall scale feature. The higher the first satellite radar confidence score, the higher the consistency between the rockfall scale feature and the local historical rockfall scale feature, that is, the higher the reliability of the rockfall scale feature. For example, the similarity between the rockfall scale feature and the local historical rockfall scale feature can be calculated using methods such as cosine similarity, Euclidean distance, and relative deviation amplitude. Taking the relative deviation amplitude as an example, if the local historical rockfall scale feature has a diameter of 0.8 meters and the rockfall scale feature has a diameter of 0.7 meters, the relative deviation amplitude between the two is (0.8 - 0.7) / 0.8 = 0.125. Since the essence of the relative deviation amplitude is the degree of difference, the smaller the relative deviation amplitude, the smaller the difference between the rockfall scale feature and the local historical rockfall scale feature. Therefore, the first satellite radar confidence score is 1 - 0.125 = 0.875. In this way, by reverse-engineering the historical rockfall scale patterns to verify the consistency of the current rockfall scale characteristics, the interference of random errors in a single monitoring session on the credibility assessment of satellite data is effectively reduced, making the confidence level of the first satellite radar more consistent with the actual scenario.

[0071] Next, the mean of the rockfall scale feature and the local historical rockfall scale feature is calculated to obtain the fused local rockfall scale feature, where the fused local rockfall scale feature = (rockfall scale feature + local historical rockfall scale feature) / 2. For example, if the local historical rockfall scale feature has a diameter of 0.8 meters and the rockfall scale feature has a diameter of 0.7 meters, then the fused local rockfall scale feature = (0.8 + 0.7) / 2 = 0.75. The fused local rockfall scale feature integrates current and historical information to obtain a more stable local rockfall scale benchmark, avoiding the bias of data from a single time point.

[0072] Furthermore, the minimum resolution of the first radar data is obtained, which is the minimum resolution of the satellite radar. The minimum resolution refers to the smallest physical size of an object that the satellite radar can stably distinguish and accurately characterize under current observation conditions (usually expressed in meters). For example, the minimum resolution of commonly used synthetic aperture radar (SAR) satellites is often 1 meter, meaning that the satellite may not be able to form clear echo characteristics for objects with a diameter less than 1 meter (such as small falling rocks), easily leading to blurred identification or missed detection. The minimum resolution of the first radar data directly determines the satellite radar's ability to identify falling rocks of different sizes.

[0073] Furthermore, the ratio of the fused local rockfall scale features to the minimum resolution is calculated to obtain the second satellite radar confidence score. The second satellite radar confidence score = fused local rockfall scale features / minimum resolution. The maximum second satellite radar confidence score is 1. This is because if the ratio of the fused local rockfall scale features to the minimum resolution is ≥1, it indicates that the actual size of the rockfall is greater than or equal to the minimum size that the satellite radar can clearly identify, and the satellite radar has the ability to stably identify the rockfall. In this case, the confidence score takes the maximum value of 1. If the ratio of the fused local rockfall scale features to the minimum resolution is <1, then this ratio is used as the second satellite radar confidence score. The second satellite radar confidence score is positively correlated with the satellite radar's recognition ability. For example, if the fused local rockfall scale features are 0.75 and the minimum resolution is 1 meter, then the second satellite radar confidence score = 0.75 / 1 = 0.75. The second satellite radar confidence score can quantify the satellite radar's ability to identify rockfall scale features. The smaller the rockfall scale features are compared to the minimum resolution of the satellite radar, the greater the recognition error, and the lower the second satellite radar confidence score.

[0074] Finally, the satellite radar confidence score is calculated based on the first and second satellite radar confidence scores. Optionally, the satellite radar confidence score = (first satellite radar confidence score + second satellite radar confidence score) / 2. For example, if the first satellite radar confidence score is 0.875 and the second satellite radar confidence score is 0.75, then the satellite radar confidence score = (0.875 + 0.75) / 2 = 0.8125. The satellite radar confidence score is a quantitative indicator between 0 and 1, combining the first and second satellite radar confidence scores. It comprehensively evaluates the reliability of satellite data from two dimensions: consistency with historical patterns and the physical constraints of the actual identification capability of satellite radar. This provides an accuracy reference for subsequent use of satellite data to identify the scale of rockfalls across the entire area. The higher the satellite radar confidence score, the more reliable the identification results of the satellite data.

[0075] Furthermore, the confidence analysis module 13 in the system also includes:

[0076] The scale recognition model calling unit is used to obtain the rockfall scale recognizer, wherein the rockfall scale recognizer is obtained by training the sample first radar data set and the sample rockfall scale feature set;

[0077] The rockfall scale identification unit is used to input the first radar data into the rockfall scale identifier and identify and output the rockfall scale characteristics, wherein the rockfall scale characteristics include the number of rocks.

[0078] In this embodiment, a rockfall scale identifier is first obtained. This identifier is trained using a sample first radar data set and a sample rockfall scale feature set. For example, from historically collected radar data from satellite radar, several valid radar data points covering different scenarios (such as different seasons, weather conditions, and terrains) are selected to form a sample first radar data set. For each segment of radar data in this set, rockfall scale features, such as the number of rocks, are manually labeled using a combination of manual interpretation and on-site verification, forming a sample rockfall scale feature set that corresponds one-to-one with the sample first radar data set.

[0079] For example, the rockfall scale recognizer can be trained using the following technical path: 1. Data preparation: Divide the sample first radar data set and the sample rockfall scale feature set into training set, validation set, and test set according to a ratio of 7:1.5:1.5. 2. Model construction: Optionally, a lightweight CNN architecture can be used to construct the rockfall scale recognizer, which mainly consists of an input layer, a feature extraction module, and an output layer. The input layer receives the sample first radar data and retains the grayscale difference features between the rockfall and the background. The feature extraction module contains three progressive convolutional blocks, each containing a 3×3 convolutional layer (increasing the number of channels from 32 to 64 to 128), batch normalization, ReLU activation function, and 2×2 max pooling. Local pixel associations are captured by sliding convolutional kernels, batch normalization stabilizes the feature distribution, ReLU enhances nonlinear expression, and the pooling layer gradually compresses the spatial dimension (256×256→128×128→64 ...64×256→64×128→64×256→64×256→64×256→ The process transforms the 8×8×128 feature map into a 128-dimensional feature vector, effectively extracting key features such as strong reflective point clusters and densely distributed textures of fallen rocks. The feature extraction module also includes a global average pooling layer, which transforms the 8×8×128 feature map into a 128-dimensional feature vector. It compresses redundant information through mean aggregation while retaining the global statistical features of the rock distribution. The output layer consists of a fully connected network with 64 neurons and a linear output layer. The fully connected network integrates feature vector information through nonlinear mapping, and the linear output layer directly outputs the predicted number of fallen rocks in a single-kilometer sub-region, realizing a direct mapping from features to scale parameters. 3. Model Training: The first radar data of the samples in the training set is used as the input feature, and the corresponding sample rockfall scale feature is used as the supervision label. The mean squared error (MSE) loss function is selected to adapt to the continuous regression requirements of rockfall quantity. The optimizer adopts the basic Adam optimizer (initial learning rate 0.001, no weight decay). The parameter tuning is simplified by fixing the learning rate, which reduces the training complexity. During the iteration process, 64 samples are input in each iteration (increasing the batch size to improve the statistical stability of features). The convolution kernel weights and fully connected layer parameters are updated by the backpropagation algorithm to continuously minimize the MSE of the predicted value and the labeled value. The performance is evaluated with the validation set every 5 rounds. When the difference between the MSE of the training set and the validation set is ≤3, it is considered to have converged, and the rockfall scale recognizer is obtained after training.

[0080] Secondly, the first radar data is input into a rockfall scale identifier, which identifies and outputs rockfall scale characteristics, including the number of rocks. For example, the first radar data is input into a pre-trained rockfall scale identifier. The identifier analyzes the distribution pattern of the radar echoes and outputs rockfall scale characteristics including the number of rocks (e.g., 5 rocks per kilometer). This achieves automated statistics on the number of rocks within the highway area, obtaining macroscopic distribution information of rocks across the entire highway, providing data support for overall disaster risk assessment.

[0081] In summary, compared to existing technologies, this application analyzes the satellite radar confidence level of the first radar data based on the aforementioned rockfall scale characteristics and local historical radar data of a specified local area, and identifies the rockfall scale characteristics within the highway area to obtain the rockfall scale features. Thus, the reliability of satellite data is quantified through satellite radar confidence level, and automated statistics of the number of rockfalls across the entire area are achieved, providing reliable data support for subsequent disaster early warning.

[0082] The data processing output module 14 is used to analyze and obtain the identification confidence level based on the rockfall scale characteristics, rockfall size characteristics and historical radar data of the field, calculate the rockfall confidence level in combination with the satellite radar confidence level, analyze and compensate for the rockfall disaster rate, and use it as the monitoring data processing result.

[0083] Historical patterns of average rockfall scale and size in highway field statistics can be used to assess the reliability of current rockfall scale and size characteristics. When rockfall scale and size characteristics deviate significantly from historical patterns, it may be due to errors in the data collection or identification process, such as scale misjudgment caused by instantaneous scanning angle deviation of UAVs, or scale counting deviation caused by satellite radar interference from severe weather. Therefore, directly using rockfall scale and size characteristics for disaster assessment may lead to over-warning and overestimation of risk.

[0084] To address the aforementioned issues, this application analyzes and obtains the identification confidence level based on the rockfall scale characteristics, rockfall size characteristics, and historical radar data of the field. It then calculates the rockfall confidence level by combining the satellite radar confidence level with the rockfall confidence level, analyzes and compensates for the rockfall disaster rate, and uses this as the result of monitoring data processing.

[0085] Specifically, the data processing output module 14 in the system includes:

[0086] The historical feature extraction unit is used to acquire radar data of the highway field over a historical period, obtain historical radar data of the field, and extract the average rockfall scale feature and average rockfall size feature identified by the historical radar data of the field as historical field rockfall scale feature and historical field rockfall size feature.

[0087] The identification confidence calculation unit is used to calculate the average similarity between the rockfall scale feature, the rockfall size feature and the historical field rockfall scale feature and the historical field rockfall size feature, and to obtain the identification confidence.

[0088] The rockfall disaster rate decision unit is used to input the rockfall scale characteristics and rockfall size characteristics into the highway disaster classifier, and make a decision classification to obtain the rockfall disaster rate.

[0089] The disaster early warning output unit is used to calculate the rockfall confidence level based on the satellite radar confidence level and the identification confidence level, perform compensation calculation on the rockfall disaster rate to obtain the compensated rockfall disaster rate, determine whether it is greater than or equal to the disaster rate threshold, and issue a disaster early warning.

[0090] In this embodiment, radar data from the highway field over a historical period is first acquired to obtain historical radar data of the field. The average rockfall scale characteristics (e.g., the average diameter of historical rocksfall is 0.8 meters) and average rockfall size characteristics (e.g., the historical average number of rocksfall per kilometer is 6) of the historical radar data of the field are extracted as historical rockfall scale characteristics and historical rockfall size characteristics. In this way, a long-term benchmark for the characteristics of rockfall in the highway field is established to measure the consistency between the current monitoring data and historical patterns, and to avoid the random deviation of data in a single time period.

[0091] Secondly, the similarity between the rockfall scale characteristics and the historical rockfall scale characteristics, and the similarity between the rockfall size characteristics and the historical rockfall size characteristics are calculated. Then, the mean similarity is calculated to obtain the recognition confidence level. Optionally, the relative deviation amplitude can be calculated first, and then the similarity can be obtained by subtracting the relative deviation amplitude from 1. For example, if the current rockfall scale characteristic is a diameter of 0.75 meters, the historical rockfall scale characteristic is an average diameter of 0.8 meters, the rockfall size characteristic is 5 rocks per kilometer, and the historical rockfall size characteristic is 6 rocks per kilometer, then the similarity between the rockfall scale characteristic and the historical rockfall scale characteristic is 1 - (|0.75 - 0.8| / 0.8) ≈ 0.94, and the similarity between the rockfall size characteristic and the historical rockfall size characteristic is 1 - (|5 - 6| / 6) ≈ 0.83. Therefore, the identification confidence is (0.94 + 0.83) / 2 = 0.885. The identification confidence quantifies the degree of agreement between the current rockfall characteristics (including scale and size characteristics) and the historical patterns of the field. The higher the identification confidence, the more consistent the current data is with the long-term rockfall characteristics of the field.

[0092] Next, the rockfall scale feature and rockfall size feature are input into the pre-trained highway disaster classifier, and the rockfall disaster rate is obtained by decision classification. The rockfall disaster rate reflects the risk level corresponding to the current rockfall feature. For example, the rockfall scale feature (such as diameter 0.75 meters) and rockfall size feature (such as 5 rocks per kilometer) are input into the pre-trained highway disaster classifier, and the rockfall disaster rate (such as 0.6) is obtained by decision classification.

[0093] For example, a highway disaster classifier can be trained using the following technical approach: 1. Data preparation: Collect complete data on past rockfall disaster cases in the highway field, obtain scale features such as the average diameter of the rocks in each case, and scale features such as the number of rocks per kilometer. Combined with on-site survey records, quantify the actual disaster impact (such as the degree of road damage, the duration of traffic interruption, and the extent of facility damage) into rockfall disaster rate labels ranging from 0% to 100%. For example, when a single rock with a diameter of 1.5 meters causes partial deformation of the guardrail but does not affect traffic, the rockfall disaster rate can be labeled as 15%. However, when 10 rocks with a diameter of more than 2 meters cause a full-width highway interruption for 8 hours, the rockfall disaster rate can be labeled as 85%. These data are then divided into training and testing sets in an 8:2 ratio. 2. Model Construction: Optionally, a gradient boosting tree (such as XGBoost) architecture is used to construct the highway disaster classifier, mainly composed of an input layer, decision trees, and an output layer. The input layer receives rockfall scale features and rockfall size features. 100 decision trees form a weak classifier set. By learning the local correlation between rockfall scale features, rockfall size features, and rockfall disaster rate layer by layer, the prediction accuracy is continuously optimized. The output layer generates the final rockfall disaster rate by weighted integration of the decision results of all trees. 3. Model Training: With mean squared error (MSE) as the optimization objective, the initial learning rate is set to 0.1 and the tree depth to 5 to balance fitting and generalization. The parameters are dynamically adjusted through 5-fold cross-validation. When the validation set MSE does not decrease for 5 consecutive rounds, the learning rate is reduced to 0.01 and the tree depth is increased to 7. At the same time, an early stopping mechanism is introduced, and training is terminated when the training set MSE < 5% and the validation set MSE < 8%, resulting in a trained highway disaster classifier.

[0094] Finally, based on the satellite radar confidence score and the identification confidence score, the rockfall confidence score is calculated. A compensation calculation is then performed on the rockfall disaster rate to obtain the compensated rockfall disaster rate. It is then determined whether this rate is greater than or equal to a disaster rate threshold, and a disaster warning is issued. Specifically, the rockfall confidence score is calculated as (satellite radar confidence score + identification confidence score) / 2. This score integrates the satellite radar confidence score (reflecting the actual ability of satellite radar to identify rockfalls) and the identification confidence score (reflecting the consistency between current rockfall characteristics and historical patterns). It is a comprehensive quantification of current data quality. A higher rockfall confidence score indicates that the satellite radar data has a more reliable physical identification capability in the current scenario, and that the scale and characteristics of the currently monitored rockfalls are more consistent with long-term historical patterns, with less influence from accidental errors or abnormal interference. The compensated rockfall disaster rate is calculated as follows: Compensated Rockfall Disaster Rate = Rockfall Confidence Level × Rockfall Disaster Rate. The rockfall disaster rate is compensated based on the rockfall confidence level. A higher confidence level results in a more accurate and closer compensated rate to the original rate. Conversely, when the confidence level is low due to factors such as satellite imagery ambiguity, the disaster rate is adjusted downwards. This effectively reduces risk overestimation caused by unreliable data, resulting in a compensated rate that better reflects the true risk level of the current scenario. The compensated disaster rate is then used as the processed result of highway monitoring data and sent to relevant personnel, such as highway maintenance staff, for reference.

[0095] Furthermore, the disaster rate threshold of rockfall can be used for judgment. The disaster rate threshold can be dynamically set according to the actual operation scenario of the highway. For example, for trunk highways with an average daily traffic volume of more than 10,000 vehicles or steep slope sections with high incidence of geological disasters, the disaster rate threshold can be set to 50%. For branch highways with low traffic volume, the disaster rate threshold can be relaxed to 70%. Those skilled in the art can flexibly adjust it in combination with factors such as highway grade, traffic density, and sensitivity of the surrounding environment. Finally, the compensated rockfall disaster rate is compared with the disaster rate threshold. When the compensated rockfall disaster rate is greater than or equal to the disaster rate threshold, a disaster warning is issued to obtain the monitoring data processing results.

[0096] For example, if the satellite radar confidence level is 0.8125, the identification confidence level is 0.885, and the rockfall disaster rate is 0.6, then the rockfall confidence level = (0.8125 + 0.885) / 2 = 0.849, and the compensated rockfall disaster rate = 0.849 × 0.6 = 0.51. If the disaster rate threshold is 50%, since 0.51 > 50%, a disaster warning is triggered. In this way, by correcting the original disaster rate through data reliability, misjudgment of risk due to data bias (such as satellite misjudgment, short-term abnormal data, etc.) is avoided, making the disaster rate more consistent with the actual scenario.

[0097] In summary, compared to existing technologies, this application analyzes and obtains the identification confidence level based on the aforementioned rockfall scale characteristics, rockfall size characteristics, and historical radar data of the field. It then calculates the rockfall confidence level by combining this with the satellite radar confidence level, analyzes and compensates for the rockfall disaster rate, and uses this as the result of monitoring data processing. Thus, by compensating for the rockfall disaster rate with the rockfall confidence level and combining this with a disaster rate threshold, accurate disaster early warning can be achieved, improving the reliability of highway disaster early warning decisions.

[0098] In summary, the embodiments of this application have at least the following technical effects:

[0099] Compared to existing technologies, this application first uses satellite radar to collect first radar data within the highway area, and then uses UAV radar to collect second radar data for a designated local area within the highway area. This achieves both global coverage and detailed local data collection, providing a reliable data foundation for subsequent rockfall feature identification, confidence analysis, and early warning decision-making.

[0100] Secondly, based on the second radar data, this application identifies the scale characteristics of falling rocks within the highway area, obtaining the scale features of the falling rocks. Thus, by intelligently processing the second radar data of a designated local area collected by the UAV radar using a deep learning model, the scale features of falling rocks are accurately extracted, solving the problems of low efficiency and large errors in traditional manual identification, and providing reliable data support for subsequent disaster early warning decisions.

[0101] Furthermore, this application analyzes the satellite radar confidence level of the first radar data based on the aforementioned rockfall scale characteristics and local historical radar data of a specified local area, and identifies the rockfall scale characteristics within the highway area to obtain the rockfall scale characteristics. In this way, the reliability of satellite data is quantified through satellite radar confidence level, and automated statistics of the number of rockfalls across the entire area are achieved, providing reliable data support for subsequent disaster early warning.

[0102] Finally, based on the aforementioned rockfall scale characteristics, rockfall size characteristics, and historical radar data of the area, this application analyzes and obtains the identification confidence level. Combined with the satellite radar confidence level, it calculates the rockfall confidence level, analyzes and compensates for the rockfall disaster rate, and uses this as the monitoring data processing result. Thus, by compensating for the rockfall disaster rate with the rockfall confidence level and combining it with a disaster rate threshold, accurate disaster early warning can be achieved, improving the reliability of highway disaster early warning decisions.

[0103] Through the aforementioned technical solution, this application achieves full coverage of highway monitoring data and detailed collection of high-risk designated local areas through collaborative perception of satellite radar and UAV radar. This enhances the comprehensiveness and relevance of data collection. Based on an artificial intelligence model, it accurately identifies the scale and size characteristics of rockfalls. Combining satellite radar confidence and identification confidence with dual-dimensional verification, it dynamically compensates for rockfall disaster rates through rockfall confidence. Finally, based on disaster rate thresholds, it achieves accurate disaster early warning. This effectively overcomes the limitations of single monitoring methods, improves the reliability and accuracy of dynamic highway disaster early warning, and can fully adapt to the safety protection needs of complex highway areas such as mountainous regions and canyons.

[0104] Example 2, as Figure 2 As shown in the figure, this embodiment of the invention also provides a method for processing radar-sensed highway monitoring data, including:

[0105] First radar data within the highway area is collected using satellite radar, and second radar data within a designated local area within the highway area is collected using drone radar.

[0106] Based on the second radar data, the scale characteristics of falling rocks in the highway area are identified to obtain the scale characteristics of falling rocks.

[0107] Based on the rockfall scale characteristics and local historical radar data of the specified local area, the satellite radar confidence level of the first radar data is analyzed, and the rockfall scale characteristics in the highway area are identified to obtain the rockfall scale characteristics.

[0108] Based on the characteristics of rockfall scale, rockfall size, and historical radar data of the field, the identification confidence level is obtained through analysis. The rockfall confidence level is calculated by combining the satellite radar confidence level. The rockfall disaster rate is then analyzed and compensated, which serves as the result of the monitoring data processing.

[0109] Specifically, the phrase "collecting first radar data within the highway area via satellite radar, and collecting second radar data within a designated local area of ​​the highway area via UAV radar" includes:

[0110] First radar data within the highway area is collected using satellite radar.

[0111] A designated local area is defined within the highway field, wherein the designated local area is a part of the highway field;

[0112] The second radar data of the designated local area is collected using the drone radar.

[0113] Specifically, the step of "identifying the scale features of falling rocks within the highway area based on the second radar data to obtain the scale features of the falling rocks" includes:

[0114] Obtain a rockfall scale identifier;

[0115] The second radar data is input into the rockfall scale identifier, and the rockfall scale characteristics are obtained by the identification output.

[0116] Furthermore, the "rockfall scale identifier" includes:

[0117] Based on historical data of falling rocks monitored by UAV radar, a sample second radar data set was collected, and all falling rock scales in each sample second radar data set were labeled to obtain a sample falling rock scale feature set.

[0118] Construct a rockfall scale identifier based on deep learning, where the input data is second radar data and the output data is rockfall scale features;

[0119] Using the sample second radar data set and the sample rockfall scale feature set, the rockfall scale recognizer is trained and optimized in a supervised manner. After verifying that convergence is satisfied, the training is completed and the rockfall scale recognizer is obtained.

[0120] Specifically, the step of "analyzing the satellite radar confidence level of the first radar data based on the rockfall scale characteristics and local historical radar data of the specified local area" includes:

[0121] The radar data within a specified local area over a historical period is obtained as local historical radar data. The historical rockfall scale feature set of the local historical radar data is extracted and the mean is calculated to obtain the local historical rockfall scale features.

[0122] Calculate the similarity between the rockfall scale features and the local historical rockfall scale features to obtain the first satellite radar confidence score;

[0123] Calculate the mean of the rockfall scale features and the local historical rockfall scale features to obtain the fused local rockfall scale features;

[0124] The minimum resolution for obtaining the first radar data;

[0125] The ratio of the fused local rockfall scale features to the minimum resolution is calculated to obtain the second satellite radar confidence score, wherein the maximum second satellite radar confidence score is 1.

[0126] The satellite radar confidence level is calculated based on the confidence levels of the first and second satellite radars.

[0127] Furthermore, the phrase "identifying the scale characteristics of falling rocks within the highway area and obtaining the scale characteristics of falling rocks" includes:

[0128] A rockfall scale identifier is obtained, wherein the rockfall scale identifier is trained using a sample first radar data set and a sample rockfall scale feature set;

[0129] The first radar data is input into the rockfall scale identifier, and the identifier outputs the rockfall scale characteristics, wherein the rockfall scale characteristics include the number of rocks.

[0130] Specifically, the phrase "based on the rockfall scale characteristics, rockfall size characteristics, and historical radar data of the field, analyze and obtain the identification confidence level, combine the satellite radar confidence level to calculate the rockfall confidence level, analyze and obtain the rockfall disaster rate, and perform disaster early warning judgment" includes:

[0131] The radar data of the highway field during the historical period is obtained, the historical radar data of the field is obtained, and the average rockfall scale feature and average rockfall size feature identified by the historical radar data of the field are extracted as the historical field rockfall scale feature and historical field rockfall size feature.

[0132] Calculate the average similarity between the rockfall scale features and rockfall size features and the historical field rockfall scale features and historical field rockfall size features to obtain the identification confidence level;

[0133] The rockfall scale features and rockfall size features are input into the highway disaster classifier, and the rockfall disaster rate is obtained by decision classification.

[0134] Based on the satellite radar confidence level and identification confidence level, the rockfall confidence level is calculated, the rockfall disaster rate is compensated, the compensated rockfall disaster rate is obtained, and it is determined whether it is greater than or equal to the disaster rate threshold to issue a disaster warning.

[0135] In summary, the embodiments of this application have at least the following technical effects:

[0136] Compared to existing technologies, this application firstly collects first radar data within the highway area using satellite radar, and then collects second radar data for a designated local area within the highway area using UAV radar. This achieves both global coverage and detailed local data collection, providing a reliable data foundation for subsequent rockfall feature identification, confidence analysis, and early warning decision-making. Secondly, based on the second radar data, the scale characteristics of rockfalls within the highway area are identified, obtaining rockfall scale features. A deep learning model is used to intelligently process the second radar data for the designated local area collected by the UAV radar, accurately extracting rockfall scale features. This solves the problems of low efficiency and large errors in traditional manual identification, providing reliable data support for subsequent disaster early warning decision-making. Thirdly, based on the rockfall scale features and local historical radar data for the designated local area, the satellite radar confidence level of the first radar data is analyzed, and rockfall scale features are identified, obtaining rockfall scale features. The reliability of satellite data is quantified through satellite radar confidence measurement, and automated statistics of the number of rockfalls across the entire area are achieved, providing reliable data support for subsequent disaster early warning. Finally, based on the characteristics of rockfall scale and size, and historical radar data of the area, the identification confidence level is analyzed and obtained. Combined with the satellite radar confidence level, the rockfall confidence level is calculated. The rockfall disaster rate is then analyzed and compensated, and used as the result of monitoring data processing. By compensating the rockfall disaster rate with the rockfall confidence level and combining it with a disaster rate threshold, accurate disaster early warning can be achieved, improving the reliability of highway disaster early warning decisions. This effectively overcomes the limitations of single monitoring methods, improves the reliability and accuracy of dynamic field highway disaster early warning, and can fully adapt to the safety protection needs of complex highway areas such as mountainous regions and canyons.

[0137] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0138] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0140] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0142] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0143] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A radar-based highway monitoring data processing system, characterized in that, The system includes: The multi-source data acquisition module is used to acquire first radar data within the highway area via satellite radar and second radar data within a specified local area within the highway area via UAV radar. The rockfall scale identification module is used to identify the scale characteristics of rocksfalling within the highway area based on the second radar data, and to obtain the rockfall scale characteristics. The confidence analysis module is used to analyze the satellite radar confidence of the first radar data based on the rockfall scale characteristics and local historical radar data of the specified local area, and to identify the rockfall scale characteristics in the highway area to obtain the rockfall scale characteristics. The data processing output module is used to analyze and obtain the identification confidence level based on the rockfall scale characteristics, rockfall size characteristics and historical radar data of the field, calculate the rockfall confidence level by combining the satellite radar confidence level, analyze and compensate for the rockfall disaster rate, and use it as the monitoring data processing result.

2. The radar-sensing highway monitoring data processing system according to claim 1, characterized in that, The multi-source data acquisition module includes: The first data acquisition unit is used to acquire first radar data within the highway area via satellite radar; A local field determination unit is used to determine a specified local field within a highway field, wherein the specified local field is a part of the highway field; The second data acquisition unit is used to acquire second radar data of the designated local area through the UAV radar.

3. The radar-sensing highway monitoring data processing system according to claim 1, characterized in that, The rockfall scale recognition module includes: The scale recognition model calling unit is used to obtain the rockfall scale recognizer; The rockfall scale identification unit is used to input the second radar data into the rockfall scale identifier and identify and output the rockfall scale features.

4. The radar-sensing highway monitoring data processing system according to claim 3, characterized in that, The scale recognition model invocation unit includes: The data preparation unit is used to collect sample second radar data sets based on historical data of UAV radar monitoring of falling rocks, and to label all falling rock scales in each sample second radar data set to obtain sample falling rock scale feature set. The model building unit is used to build a rockfall scale recognizer based on deep learning, wherein the input data is the second radar data and the output data is the rockfall scale features. The model training unit is used to perform supervised training and optimization of the rockfall scale recognizer using the sample second radar data set and the sample rockfall scale feature set. After verifying that convergence is satisfied, the training is completed and the rockfall scale recognizer is obtained.

5. The radar-sensing highway monitoring data processing system according to claim 1, characterized in that, The confidence analysis module includes: The historical scale feature extraction unit is used to acquire radar data within a specified local field during a historical period, as local historical radar data, extract the historical rockfall scale feature set of the local historical radar data, and calculate the mean to obtain local historical rockfall scale features. The first satellite radar confidence calculation unit is used to calculate the similarity between the rockfall scale features and the local historical rockfall scale features to obtain the first satellite radar confidence. The fusion scale feature acquisition unit is used to calculate the mean of the rockfall scale feature and the local historical rockfall scale feature to obtain the fused local rockfall scale feature. Minimum resolution acquisition unit, used to acquire the minimum resolution of the first radar data; The second satellite radar confidence calculation unit is used to calculate the ratio of the fused local rockfall scale features to the minimum resolution to obtain the second satellite radar confidence, wherein the maximum second satellite radar confidence is 1. The satellite radar confidence calculation unit is used to calculate the satellite radar confidence based on the first satellite radar confidence and the second satellite radar confidence.

6. The radar-sensing highway monitoring data processing system according to claim 1, characterized in that, The confidence analysis module further includes: The scale recognition model calling unit is used to obtain the rockfall scale recognizer, wherein the rockfall scale recognizer is obtained by training the sample first radar data set and the sample rockfall scale feature set; The rockfall scale identification unit is used to input the first radar data into the rockfall scale identifier and identify and output the rockfall scale characteristics, wherein the rockfall scale characteristics include the number of rocks.

7. The radar-sensing highway monitoring data processing system according to claim 1, characterized in that, The data processing output module includes: The historical feature extraction unit is used to acquire radar data of the highway field over a historical period, obtain historical radar data of the field, and extract the average rockfall scale feature and average rockfall size feature identified by the historical radar data of the field as historical field rockfall scale feature and historical field rockfall size feature. The identification confidence calculation unit is used to calculate the average similarity between the rockfall scale feature, the rockfall size feature and the historical field rockfall scale feature and the historical field rockfall size feature, and to obtain the identification confidence. The rockfall disaster rate decision unit is used to input the rockfall scale characteristics and rockfall size characteristics into the highway disaster classifier, and make a decision classification to obtain the rockfall disaster rate. The disaster early warning output unit is used to calculate the rockfall confidence level based on the satellite radar confidence level and the identification confidence level, perform compensation calculation on the rockfall disaster rate to obtain the compensated rockfall disaster rate, determine whether it is greater than or equal to the disaster rate threshold, and issue a disaster early warning.

8. A method for processing radar-sensing highway monitoring data, characterized in that, include: First radar data within the highway area is collected using satellite radar, and second radar data within a designated local area within the highway area is collected using drone radar. Based on the second radar data, the scale characteristics of falling rocks in the highway area are identified to obtain the scale characteristics of falling rocks. Based on the rockfall scale characteristics and local historical radar data of the specified local area, the satellite radar confidence level of the first radar data is analyzed, and the rockfall scale characteristics in the highway area are identified to obtain the rockfall scale characteristics. Based on the characteristics of rockfall scale, rockfall size, and historical radar data of the field, the identification confidence level is obtained through analysis. The rockfall confidence level is calculated by combining the satellite radar confidence level. The rockfall disaster rate is then analyzed and compensated, which serves as the result of the monitoring data processing.

Images