A road exploration and evaluation method, system and storage medium based on multidimensional data

Through the multi-dimensional data road exploration and evaluation method, combined with underground radar maps and road surface images, the data augmentation and air-removing detection model are used to generate road evaluation maps, which solves the problems of incomplete information acquisition and low efficiency in traditional methods and realizes road health Comprehensive analysis of status and optimized maintenance plan.

CN118707516BActive Publication Date: 2024-11-05HUBEI COMM PLANNING & DESIGN INST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411195719.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-05
Estimated Expiration
2044-08-29

AI Technical Summary

Technical Problem

Traditional road exploration methods rely on visual inspection and manual measurement, and cannot fully obtain underground structural information, are susceptible to subjective factors, lack consistency and accuracy of identification and evaluation, and are inefficient in data processing, making it difficult to generate intuitive decision support.

Method used

Using a multi-dimensional data-based road exploration and evaluation method, the underground radar map is obtained through ground penetrating radar, preprocessing and feature extraction is performed, and the data augmentation model and air-removing detection model are combined to integrate underground and road surface information, and the road evaluation map is generated using GIS technology. .

Benefits of technology

It has achieved comprehensive acquisition of information at different levels of roads, improved data quality and model generalization capabilities, provided comprehensive road health status analysis, generated intuitive maps, supported decision makers to quickly understand and optimize maintenance plans, extend the life of the road, and reduce driving risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118707516B_ABST
    Figure CN118707516B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of road disease identification, specifically to a road exploration and evaluation method, system and storage medium based on multidimensional data, including: obtaining underground radar maps of multiple underground locations of a target road based on a ground penetrating radar device, preprocessing the underground radar maps to obtain standard underground radar maps, and constructing a training set based on the standard underground radar maps of multiple underground locations of the target road. The present invention can comprehensively obtain information on different levels of the road, including underground structure and road surface status, by combining underground radar maps and road surface images, and preprocessing and standardizing underground radar maps and road surface images, which helps to improve data quality and comparability, making subsequent analysis and model training more effective and accurate.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of road damage identification, and in particular to a road exploration and evaluation method, system and storage medium based on multidimensional data. Background Art

[0002] The purpose of road survey is to evaluate the structural integrity, geological conditions, etc. of the road to ensure the safety and durability of the road.

[0003] Traditional methods usually rely on visual inspection and manual measurement, which cannot fully obtain the underground structural information of the road. Problems such as underground cavities and cracks may be hidden under the surface and difficult to detect through visual inspection. In addition, under traditional methods, the identification of road surface diseases relies on manual observation, which is easily affected by subjective factors and the skill level of the operator, resulting in insufficient consistency and accuracy in identification and evaluation. In addition, under traditional methods, the efficiency of processing large amounts of data and generating comprehensive maps is inefficient, and often relies on manual drawing and recording, which is time-consuming and prone to errors. In addition, the reports and records generated by traditional methods may lack clear information expression and visualization, making it difficult to provide decision makers with intuitive understanding and rapid decision support. Summary of the invention

[0004] The purpose of the present invention is to address the problems existing in the background technology and to propose a road exploration and evaluation method, system and storage medium based on multidimensional data.

[0005] The technical solution of the present invention is a road exploration and evaluation method based on multidimensional data, comprising:

[0006] Acquire underground radar maps of multiple underground locations of the target road based on a ground penetrating radar device, perform preprocessing operations on the underground radar maps to obtain standard underground radar maps, and construct a training set based on the standard underground radar maps of multiple underground locations of the target road;

[0007] Performing data augmentation on the training set by using a trained data augmentation model to obtain an augmented training set, constructing a gap detection model, and training the gap detection model based on the augmented training set to obtain a trained gap detection model;

[0008] Performing a void detection on multiple underground locations of the target road using the trained void detection model to obtain void results of the multiple underground locations of the target road;

[0009] Acquire road surface images at multiple locations of a target road based on an image acquisition device, perform preprocessing operations on the road surface images to obtain standard road surface images, perform feature extraction on the standard road surface images to obtain an image feature set, and perform road surface disease detection on the image feature set using a trained support vector machine-based road surface disease detection model to obtain road surface disease results at multiple locations of the target road;

[0010] Integrate the void results of multiple underground locations and the pavement damage results of multiple locations of the target road to obtain a data attribute list, wherein the data attribute list includes an underground radar map number, a pavement image number, longitude and latitude information, void results, and pavement damage results;

[0011] A road map of the target road is obtained, and underground radar maps of multiple underground locations and road surface images of multiple locations of the target road are imported into the road map through GIS technology to obtain a road association map, and the data attribute list is associated with the road association map to obtain a road evaluation map.

[0012] Preferably, the underground radar map is preprocessed to obtain a standard underground radar map, including:

[0013] The underground radar spectrum is subjected to static correction, and the static correction formula is as follows:

[0014] f(t)=x(t-t0);

[0015] Wherein, f(t) represents the underground radar spectrum after static correction, x(t-t0) represents the amplitude in the underground radar spectrum at the time (t-t0), t0 represents the time of left translation, and t represents the time at the maximum amplitude position in the underground radar spectrum;

[0016] The underground radar spectrum is subjected to a DC drift, and the DC drift formula is as follows:

[0017]

[0018] Where y′(n) represents the underground radar map after DC drift, y(n) represents the underground radar map before DC drift, and N represents the total number of sampling points in the underground radar map;

[0019] The underground radar spectrum is subjected to bandpass filtering, and the bandpass filtering formula is as follows:

[0020] Y′(ω)=Y(ω)×H(ω);

[0021] Among them, Y′(ω) represents the underground radar spectrum after bandpass filtering, Y(ω) represents the frequency domain characteristics of the underground radar spectrum, and H(ω) represents the bandpass filter.

[0022] Preferably, the data augmentation model adopts an improved ConSinGan network, the ConSinGan network includes a generator and a discriminator, the generator is used to learn the real sample distribution from the training set, and generate fake samples based on the real sample distribution, the discriminator is used to judge the fake samples, and improve the generator of the ConSinGan network through a feature learning module and a loss function, the feature learning module includes a first submodule, a second submodule and a third submodule, the first submodule is used to use a maximum pooling method to perform feature learning to obtain a first feature map, and the second submodule is used to use an average The feature learning is performed in a pooling manner to obtain a second feature map, the first feature map is point multiplied with the input feature map to obtain a third feature map, the second feature map is point multiplied with the input feature map to obtain a fourth feature map, the third feature map and the fourth feature map are feature fused to obtain a fused feature map, the third submodule is used to reduce the number of channels of the fused feature map through a convolution layer, and nonlinearly activate the fused feature map after the number of channels is reduced through a Sigmoid function to obtain an output feature map, and the output feature map is point multiplied with the input feature map to obtain a final feature map.

[0023] Preferably, the feature fusion operation formula is as follows:

[0024]

[0025] Among them, F′ represents the fusion feature map, F r represents the third feature map, F l represents the fourth feature map.

[0026] Preferably, the loss function is as follows:

[0027] minmaxl adv (G n ,D n )+αl rec (G n );

[0028] Among them, l adv represents the adversarial loss, G n and D n They represent the generator and discriminator at training stage n, α represents the artificially set weight, and l rec represents the reconstruction loss.

[0029] Preferably, the reconstruction loss formula is as follows:

[0030]

[0031] in, represents the feature map of the reconstructed output at the training stage n-1, ↑ represents the upsampling operation, Z * represents a fixed input feature map, x n Represents the feature map at training stage n.

[0032] Preferably, the void detection model adopts an improved YOLOv7-tiny model, and the YOLOv7-tiny model is improved by an enhanced attention module, the enhanced attention module includes a channel attention module and a spatial attention module, the channel attention module is used for the first spectrum features of the standard underground radar map of multiple underground locations of the target road, and the spatial attention module is used for the second spectrum features of the standard underground radar map of multiple underground locations of the target road, the first spectrum features and the second spectrum features are added element by element to obtain the target feature attention spectrum features.

[0033] Preferably, the channel attention module is expressed as follows:

[0034] M C (X)=σ(MLP(AvgPool(X))+MLP(MaxPool(X)));

[0035] Among them, M C (X) represents the first map feature, MLP represents a shared multi-layer perceptron, AvgPool represents global average pooling, MaxPool represents maximum pooling, and X represents a standard underground radar map of multiple underground locations of the target road.

[0036] Preferably, the spatial attention module is expressed as follows:

[0037] M s (X) = σ(f 7×7 (AvgPool(X));MLP(MaxPool(X));

[0038] Among them, M S (X) represents the second spectrum feature, f 7×7 Represents a 7×7 convolution operation.

[0039] Preferably, the target feature attention graph feature expression is as follows:

[0040]

[0041] Among them, X out represents the target feature attention map feature, represents element-wise multiplication, Represents element-by-element addition.

[0042] Preferably, feature extraction is performed on the standard road surface image to obtain an image feature set, including:

[0043] The texture features of the standard road surface image are extracted, and the texture feature extraction formula is as follows:

[0044]

[0045] Among them, C T represents the texture feature result, d represents the amplitude direction of the standard road image, s represents the distance, φ A represents the neighborhood of area A in the standard road image, and P represents the probability solution result;

[0046] The color features of the standard road image are extracted, and the color feature extraction formula is as follows:

[0047]

[0048] Among them, q represents the color feature, σ i and σ j They represent the brightness variance in different areas of the standard road image respectively;

[0049] The distance feature of the standard road image is extracted, and the distance feature extraction formula is as follows:

[0050]

[0051] Among them, L represents the distance of the standard road image sequence minimized and mapped to a continuous function on a straight line, F represents the continuous function, and C x and C y represents the mapping points in different areas of the standard road image, E xy Represents the proximity matrix

[0052] The texture features, color features and distance features of the standard road surface image are combined to obtain an image feature set.

[0053] Preferably, associating the data attribute list with the road association map to obtain a road assessment map comprises:

[0054] defining a display layer, wherein the display layer is used to display the void results of multiple underground locations and the road surface damage results of multiple locations of the target road;

[0055] The longitude and latitude information in the data attribute list is used as a key field and associated with a corresponding point mark on the road-related map, thereby generating the display layer on the road-related map;

[0056] Different colors and shapes are set for the road association map to distinguish between the air gap result and the road surface damage result.

[0057] The technical solution of the present invention is a road exploration and evaluation system based on multidimensional data, which is applicable to the above-mentioned road exploration and evaluation method based on multidimensional data, including:

[0058] A data processing module, the data processing module is used to obtain underground radar maps of multiple underground locations of the target road, and perform preprocessing operations to generate a standard underground radar map, and construct a training set based on the standard underground radar map for training a void detection model;

[0059] A data training module, wherein the data augmentation module is used to augment the training set through a data augmentation model to obtain an augmented training set, and use the augmented training set to train the gap detection model to obtain a trained gap detection model;

[0060] An image processing module, the image processing module is used to obtain road surface images at multiple locations of the target road, and perform preprocessing operations on these images to obtain standard road surface images, and extract features from the standard road surface images to form an image feature set;

[0061] A disease detection module, the disease detection module is used to perform road disease detection on an image feature set based on a support vector machine;

[0062] A map generation module integrates the void detection results and the pavement disease detection results into data attribute columns, and uses GIS technology to import underground radar maps, pavement images and data attribute lists into road maps to generate road assessment maps.

[0063] The technical solution of the present invention is: a computer-readable storage medium, which is applicable to the road exploration and evaluation method based on multidimensional data, comprising: at least one processor and a memory;

[0064] The memory is used to store computer-executable instructions;

[0065] The at least one processor is used to execute the computer-executable instructions stored in the memory to implement the above-mentioned road exploration and evaluation method based on multi-dimensional data.

[0066] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0067] 1. The present invention can comprehensively obtain information on different levels of roads, including underground structures and road surface conditions, by combining underground radar maps and road surface images, and preprocess and standardize underground radar maps and road surface images, which helps to improve data quality and comparability, making subsequent analysis and model training more effective and accurate. The training set is augmented using a trained data augmentation model, and then a void detection model and a road surface disease detection model are constructed, thereby improving the generalization ability and adaptability of the model and ensuring the effectiveness of the model in various practical scenarios.

[0068] 2. The present invention can provide a comprehensive analysis of the road health status by comprehensively evaluating the underground structure health and pavement usage status of the road through void detection and pavement disease detection. By combining GIS technology, underground radar maps, pavement images and related attributes are imported into road maps to generate road assessment maps, thereby providing intuitive and easy-to-understand map visualizations to help decision makers quickly analyze and understand the overall situation of the road. Moreover, based on detailed road assessment maps, road maintenance and repair plans can be accurately planned and optimized, resources can be effectively utilized, and the service life of roads can be extended. Through continuous monitoring and maintenance, the road usage experience can be improved, driving risks and vehicle damage can be reduced, and traffic fluency and efficiency can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A schematic flow chart of an overall method in an embodiment of the present invention;

[0070] Figure 2 A schematic diagram of the overall system flow in an embodiment of the present invention;

[0071] Figure 3 A schematic diagram of the overall storage medium process in an embodiment of the present invention.

[0072] Figure numerals: 1. data processing module; 2. data training module; 3. image processing module; 4. disease detection module; 5. map generation module; 6. processor; 7. memory; 8. bus. DETAILED DESCRIPTION

[0073] Embodiment 1, as Figure 1 As shown, the present invention proposes a road exploration and evaluation method based on multidimensional data, comprising:

[0074] S1. Obtain underground radar maps of multiple underground locations of a target road based on a ground penetrating radar device, perform preprocessing operations on the underground radar maps to obtain standard underground radar maps, and construct a training set based on the standard underground radar maps of multiple underground locations of the target road;

[0075] S2. Augmenting the training set with the trained data augmentation model to obtain an augmented training set, constructing a gap detection model, and training the gap detection model based on the augmented training set to obtain a trained gap detection model;

[0076] S3, performing a void detection on multiple underground locations of the target road using a trained void detection model to obtain void results of the multiple underground locations of the target road;

[0077] S4, acquiring road surface images at multiple locations of the target road based on an image acquisition device, performing preprocessing operations on the road surface images to obtain standard road surface images, performing feature extraction on the standard road surface images to obtain an image feature set, and performing road surface disease detection on the image feature set using a trained support vector machine-based road surface disease detection model to obtain road surface disease results at multiple locations of the target road;

[0078] S5. Integrate the void results of multiple underground locations and the pavement damage results of multiple locations of the target road to obtain a data attribute list, the data attribute list including the underground radar map number, the pavement image number, the longitude and latitude information, the void results and the pavement damage results;

[0079] S6. Obtain a road map of the target road, import underground radar maps of multiple underground locations and road surface images of multiple locations of the target road into the road map through GIS technology to obtain a road association map, and associate the data attribute list with the road association map to obtain a road assessment map.

[0080] In the present invention, the underground radar map is a visual representation of underground structure data obtained by ground penetrating radar equipment, showing reflection signals at different depths underground, and is used to analyze underground structures; the data augmentation model is a model for increasing the amount and diversity of training set data, and improving the performance and generalization ability of the model by transforming and expanding existing data; the void detection model is used to analyze the underground radar map and detect whether there are voids or gaps therein; the pavement disease detection model is used to analyze the characteristics of the pavement image and detect various diseases on the pavement, such as cracks, potholes, etc.; the road association map combines the underground radar map, the pavement image and its corresponding detection results (voids and diseases), and is a comprehensive road map generated by GIS technology; the road assessment map is a map obtained after a comprehensive assessment of the target road, showing the structural health status of the road and the distribution of various diseases, providing important data support for road maintenance and management.

[0081] Embodiment 2, a road exploration and evaluation method based on multidimensional data proposed by the present invention, compared with embodiment 1, this embodiment further includes: performing a preprocessing operation on the underground radar map to obtain a standard underground radar map, including:

[0082] A1. Perform static correction on underground radar spectrum. The static correction formula is as follows:

[0083] f(t)=x(t-t0);

[0084] Wherein, f(t) represents the underground radar spectrum after static correction, x(t-t0) represents the amplitude in the underground radar spectrum at the time (t-t0), t0 represents the time of left translation, and t represents the time at the maximum amplitude position in the underground radar spectrum;

[0085] A2. Perform DC drift on the underground radar spectrum. The DC drift formula is as follows:

[0086]

[0087] Where y′(n) represents the underground radar map after DC drift, y(n) represents the underground radar map before DC drift, and N represents the total number of sampling points in the underground radar map;

[0088] A3. Perform bandpass filtering on the underground radar spectrum. The bandpass filtering formula is as follows:

[0089] Y′(ω)=Y(ω)×H(ω);

[0090] Among them, Y′(ω) represents the underground radar spectrum after bandpass filtering, Y(ω) represents the frequency domain characteristics of the underground radar spectrum, and H(ω) represents the bandpass filter.

[0091] In this embodiment, static correction refers to adjusting the time axis in the underground radar spectrum so that the position of its maximum amplitude is aligned with a specific moment, usually by making it reach the maximum amplitude at a certain point on the time axis (generally zero time), so as to more clearly display the characteristics of the underground structure; DC drift refers to removing the DC component in the underground radar spectrum, usually by calculating the average value of each sampling point and adjusting the amplitude on the entire spectrum accordingly to eliminate the long-term drift effect; bandpass filtering is a signal processing method for selecting signal components within a specific frequency range in the underground radar spectrum. The bandpass filter is usually defined by a pair of cutoff frequencies, which is used to filter out low-frequency and high-frequency components and retain the intermediate frequency range to highlight the underground structure characteristics of interest.

[0092] In an optional embodiment, the data augmentation model adopts an improved ConSinGan network, which includes a generator and a discriminator. The generator is used to learn the distribution of real samples from the training set and generate fake samples based on the real sample distribution. The discriminator is used to judge the fake samples and improve the generator of the ConSinGan network through a feature learning module and a loss function. The feature learning module includes a first submodule, a second submodule and a third submodule. The first submodule is used to use a maximum pooling method to perform feature learning to obtain a first feature map, and the second submodule is used to use an average The feature learning is performed in a pooling manner to obtain a second feature map, the first feature map is point multiplied with the input feature map to obtain a third feature map, the second feature map is point multiplied with the input feature map to obtain a fourth feature map, the third feature map and the fourth feature map are feature fused to obtain a fused feature map, the third submodule is used to reduce the number of channels of the fused feature map through a convolutional layer, and nonlinearly activate the fused feature map after the number of channels is reduced through a Sigmoid function to obtain an output feature map, and the output feature map is point multiplied with the input feature map to obtain a final feature map.

[0093] It should be noted that the ConSinGan network is a generative adversarial network (GAN) model used for data augmentation. GAN consists of two parts: the generator and the discriminator, which learn and generate realistic data samples through adversarial training.

[0094] In an optional embodiment, the feature fusion operation formula is as follows:

[0095]

[0096] Among them, F′ represents the fusion feature map, F r represents the third feature map, F l represents the fourth feature map.

[0097] In an optional embodiment, the loss function is as follows:

[0098] minmaxl adv (G n ,D n )+αl rec (G n );

[0099] Among them, l adv represents the adversarial loss, G n and D n They represent the generator and discriminator at training stage n, α represents the artificially set weight, and l recrepresents the reconstruction loss.

[0100] In an optional embodiment, the reconstruction loss formula is as follows:

[0101]

[0102] in, represents the feature map of the reconstructed output at the training stage n-1, ↑ represents the upsampling operation, Z * represents a fixed input feature map, x n Represents the feature map at training stage n.

[0103] It should be noted that the adversarial loss is used to measure the adversarial performance between the generator and the discriminator. The adversarial loss is usually based on the generator's ability to fool the discriminator and the discriminator's ability to accurately classify the samples generated by the generator. The reconstruction loss is used to measure the difference or error between the generator's output and the input. The reconstruction loss is usually used to supervise the generator to generate reasonable output while maintaining the characteristics of the input data.

[0104] In an optional embodiment, the void detection model adopts an improved YOLOv7-tiny model, and improves the YOLOv7-tiny model through an enhanced attention module, the enhanced attention module includes a channel attention module and a spatial attention module, the channel attention module is used for the first spectrum features of the standard underground radar map of multiple underground locations of the target road, and the spatial attention module is used for the second spectrum features of the standard underground radar map of multiple underground locations of the target road, and the first spectrum features and the second spectrum features are added element by element to obtain the target feature attention spectrum features.

[0105] It should be noted that YOLOv7-tiny is an object detection model that evolved from the YOLO (You Only Look Once) series of models. The YOLO series of models are known for their fast and efficient features and can directly predict the category and bounding box of an object in a single forward propagation process.

[0106] In an optional embodiment, the channel attention module expression is as follows:

[0107] M C (X)=σ(MLP(AvgPool(X))+MLP(MaxPool(X)));

[0108] Among them, M C (X) represents the first map feature, MLP represents a shared multi-layer perceptron, AvgPool represents global average pooling, MaxPool represents maximum pooling, and X represents a standard underground radar map of multiple underground locations of the target road.

[0109] It should be noted that the shared multilayer perceptron is a neural network structure whose main feature is that different parts of the network share the same multilayer perceptron layer or subnetwork. This sharing greatly reduces the number of network parameters and improves the generalization ability and training efficiency of the network.

[0110] In an optional embodiment, the spatial attention module expression is as follows:

[0111] M s (X) = σ(f 7×7 (AvgPool(X));MLP(MaxPool(X)));

[0112] Among them, M S (X) represents the second spectrum feature, f 7×7 Represents a 7×7 convolution operation.

[0113] In an optional embodiment, the target feature attention map feature expression is as follows:

[0114]

[0115] Among them, X out represents the target feature attention map feature, represents element-wise multiplication, Represents element-by-element addition.

[0116] In an optional embodiment, feature extraction is performed on the standard road surface image to obtain an image feature set, including:

[0117] B1. Extract the texture features of the standard road image. The texture feature extraction formula is as follows:

[0118]

[0119] Among them, C T represents the texture feature result, d represents the amplitude direction of the standard road image, s represents the distance, φ A represents the neighborhood of area A in the standard road image, and P represents the probability solution result;

[0120] B2. Extract the color features of the standard road image. The color feature extraction formula is as follows:

[0121]

[0122] Among them, q represents the color feature, σ i and σ j They represent the brightness variance in different areas of the standard road image respectively;

[0123] B3. Extract the distance feature of the standard road image. The distance feature extraction formula is as follows:

[0124]

[0125] Among them, L represents the distance of the standard road image sequence minimized and mapped to a continuous function on a straight line, F represents the continuous function, and C x and C y represents the mapping points in different areas of the standard road image, E xy Represents the proximity matrix

[0126] B4. Combine the texture features, color features and distance features of the standard road image to obtain an image feature set.

[0127] In an optional embodiment, associating the data attribute list with the road association map to obtain a road assessment map includes:

[0128] C1. Define a display layer, which is used to display the void results of multiple underground locations and the road surface damage results of multiple locations of the target road;

[0129] C2. Associating the longitude and latitude information in the data attribute list with the corresponding point marks on the road association map as key fields, thereby generating a display layer on the road association map;

[0130] C3. Set different colors and shapes for the road association map to distinguish between air gap results and road surface damage results.

[0131] Embodiment three, as Figure 2 As shown, the present invention proposes a road exploration and evaluation system based on multidimensional data, which is applicable to a road exploration and evaluation method based on multidimensional data, including:

[0132] Data processing module 1, the data processing module 1 is used to obtain underground radar maps of multiple underground locations of the target road, and perform preprocessing operations to generate standard underground radar maps, and construct a training set based on the standard underground radar maps for training the void detection model;

[0133] Data training module 2: data augmentation module 2 is used to augment the training set through the data augmentation model to obtain an augmented training set, and use the augmented training set to train the gap detection model to obtain a trained gap detection model;

[0134] The image processing module 3 is used to obtain road surface images at multiple locations of the target road, and perform preprocessing operations on these images to obtain standard road surface images, and extract features from the standard road surface images to form an image feature set;

[0135] The disease detection module 4 is used to detect road surface diseases based on the image feature set based on the support vector machine;

[0136] Map generation module 5: Map generation module 5 integrates the void detection results and the road surface disease detection results into data attribute columns, and uses GIS technology to import underground radar maps, road surface images and data attribute lists into the road map to generate a road assessment map.

[0137] Embodiment 4, as Figure 3 As shown, a computer-readable storage medium proposed by the present invention is applicable to a road exploration and evaluation method based on multi-dimensional data, comprising: at least one processor 6 and a memory 7;

[0138] The memory 7 is used to store computer execution instructions;

[0139] At least one processor 6 is used to execute computer-executable instructions stored in the memory 7 to implement a road exploration and evaluation method based on multi-dimensional data as claimed in any one of claims 1 to 6.

[0140] In an optional embodiment, the memory 7 can be independent or integrated with the processor 6. When the memory 7 is independently provided, the device further includes a bus 8 for connecting the memory 7 and the processor 6.

[0141] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A road exploration and evaluation method based on multidimensional data, characterized in that: include: Acquire underground radar maps of multiple underground locations of the target road based on a ground penetrating radar device, perform preprocessing operations on the underground radar maps to obtain standard underground radar maps, and construct a training set based on the standard underground radar maps of multiple underground locations of the target road; Performing data augmentation on the training set by using a trained data augmentation model to obtain an augmented training set, constructing a gap detection model, and training the gap detection model based on the augmented training set to obtain a trained gap detection model; Performing a void detection on multiple underground locations of the target road using the trained void detection model to obtain void results of the multiple underground locations of the target road; Acquire road surface images at multiple locations of a target road based on an image acquisition device, perform preprocessing operations on the road surface images to obtain standard road surface images, perform feature extraction on the standard road surface images to obtain an image feature set, and perform road surface disease detection on the image feature set using a trained support vector machine-based road surface disease detection model to obtain road surface disease results at multiple locations of the target road; Integrate the void results of multiple underground locations and the pavement damage results of multiple locations of the target road to obtain a data attribute list, wherein the data attribute list includes an underground radar map number, a pavement image number, longitude and latitude information, void results, and pavement damage results; A road map of the target road is obtained, and underground radar maps of multiple underground locations and road surface images of multiple locations of the target road are imported into the road map through GIS technology to obtain a road association map, and the data attribute list is associated with the road association map to obtain a road evaluation map.

2. A road exploration and evaluation method based on multidimensional data according to claim 1, characterized in that: The underground radar map is preprocessed to obtain a standard underground radar map, including: The underground radar spectrum is subjected to static correction, and the static correction formula is as follows: f(t)=x(t-t0); Wherein, f(t) represents the underground radar spectrum after static correction, x(t-t0) represents the amplitude in the underground radar spectrum at the time (t-t0), t0 represents the time of left translation, and t represents the time at the maximum amplitude position in the underground radar spectrum; The underground radar spectrum is subjected to a DC drift, and the DC drift formula is as follows: Where y′(n) represents the underground radar map after DC drift, y(n) represents the underground radar map before DC drift, and N represents the total number of sampling points in the underground radar map; The underground radar spectrum is subjected to bandpass filtering, and the bandpass filtering formula is as follows: Y′(ω)=Y(ω)×H(ω); Among them, Y′(ω) represents the underground radar spectrum after bandpass filtering, Y(ω) represents the frequency domain characteristics of the underground radar spectrum, and H(ω) represents the bandpass filter.

3. A road exploration and evaluation method based on multidimensional data according to claim 1, characterized in that: The data augmentation model adopts an improved ConSinGan network, which includes a generator and a discriminator. The generator is used to learn the distribution of real samples from the training set and generate fake samples based on the real sample distribution. The discriminator is used to judge the fake samples and improve the generator of the ConSinGan network through a feature learning module and a loss function. The feature learning module includes a first submodule, a second submodule and a third submodule. The first submodule is used to perform feature learning using a maximum pooling method to obtain a first feature map, and the second submodule is used to perform feature learning using an average pooling method. The method performs feature learning in a manner to obtain a second feature map, performs a dot multiplication operation on the first feature map and the input feature map to obtain a third feature map, performs a dot multiplication operation on the second feature map and the input feature map to obtain a fourth feature map, performs a feature fusion operation on the third feature map and the fourth feature map to obtain a fused feature map, the third submodule is used to reduce the number of channels of the fused feature map through a convolution layer, and performs nonlinear activation on the fused feature map after the number of channels is reduced through a Sigmoid function to obtain an output feature map, and performs a dot multiplication operation on the output feature map and the input feature map to obtain a final feature map.

4. A road exploration and evaluation method based on multidimensional data according to claim 3, characterized in that: The feature fusion operation formula is as follows: Among them, F′ represents the fusion feature map, F r represents the third feature map, F l represents the fourth characteristic map; The loss function is as follows: minmaxl adv (G n ,D n )+αl rec (G n ); Among them, l adv represents the adversarial loss, G n and D n They represent the generator and discriminator at training stage n, α represents the artificially set weight, and l rec Represents the reconstruction loss, and the reconstruction loss formula is as follows: in, represents the feature map of the reconstructed output at the training stage n-1, ↑ represents the upsampling operation, Z * represents a fixed input feature map, x n Represents the feature map at training stage n.

5. The road exploration and evaluation method based on multidimensional data according to claim 1, characterized in that: The void detection model adopts an improved YOLOv7-tiny model, and improves the YOLOv7-tiny model through an enhanced attention module, wherein the enhanced attention module includes a channel attention module and a spatial attention module, wherein the channel attention module is used for a first spectrum feature of a standard underground radar map of multiple underground locations of the target road, and the spatial attention module is used for a second spectrum feature of a standard underground radar map of multiple underground locations of the target road, and the first spectrum feature and the second spectrum feature are added element by element to obtain a target feature attention spectrum feature.

6. A road exploration and evaluation method based on multidimensional data according to claim 5, characterized in that: The channel attention module expression is as follows: M C (X)=σ(MLP(AvgPool(X))+MLP(MaxPool(X))); Among them, M C (X) represents the first map feature, MLP represents a shared multi-layer perceptron, AvgPool represents global average pooling, MaxPool represents maximum pooling, and X represents a standard underground radar map of multiple underground locations of the target road; The spatial attention module expression is as follows: M S (X)=σ(f 7×7 (AvgPool(X));MLP(MaxPool(X))); Among them, M S (X) represents the second spectrum feature, f 7×7 Represents a 7×7 convolution operation; The target feature attention map feature expression is as follows: Among them, X out represents the target feature attention map feature, represents element-wise multiplication, Represents element-by-element addition.

7. A road exploration and evaluation method based on multidimensional data according to claim 1, characterized in that: Performing feature extraction on the standard road surface image to obtain an image feature set, including: The texture features of the standard road surface image are extracted, and the texture feature extraction formula is as follows: Among them, C T represents the texture feature result, d represents the amplitude direction of the standard road image, s represents the distance, φ A represents the neighborhood of area A in the standard road image, and P represents the probability solution result; The color features of the standard road image are extracted, and the color feature extraction formula is as follows: Among them, q represents the color feature, σ i and σ j They represent the brightness variance in different areas of the standard road image respectively; The distance feature of the standard road image is extracted, and the distance feature extraction formula is as follows: Among them, L represents the distance of the standard road image sequence minimized and mapped to a continuous function on a straight line, F represents the continuous function, and C x and C y represents the mapping points in different areas of the standard road image, E xy represents the proximity matrix; The texture features, color features and distance features of the standard road surface image are combined to obtain an image feature set.

8. The road exploration and evaluation method based on multidimensional data according to claim 1, characterized in that: Associating the data attribute list with the road association map to obtain a road assessment map, including: defining a display layer, wherein the display layer is used to display the void results of multiple underground locations and the road surface damage results of multiple locations of the target road; The longitude and latitude information in the data attribute list is used as a key field and associated with a corresponding point mark on the road-related map, thereby generating the display layer on the road-related map; Different colors and shapes are set for the road association map to distinguish between the air gap result and the road surface damage result.

9. A road exploration and evaluation system based on multidimensional data, which is applicable to a road exploration and evaluation method based on multidimensional data as claimed in any one of claims 1 to 8, characterized in that: include: A data processing module (1), the data processing module (1) is used to obtain underground radar maps of multiple underground locations of the target road, and perform preprocessing operations to generate a standard underground radar map, and construct a training set based on the standard underground radar map for training a void detection model; A data training module (2), wherein the data training module (2) is used to augment the training set by using a data augmentation model to obtain an augmented training set, and use the augmented training set to train a gap detection model to obtain a trained gap detection model; An image processing module (3), the image processing module (3) is used to obtain road surface images at multiple locations of the target road, and perform preprocessing operations on these images to obtain standard road surface images, and extract features from the standard road surface images to form an image feature set; A disease detection module (4), the disease detection module (4) is used to perform road disease detection on the image feature set based on a support vector machine; A map generation module (5) integrates the void detection results and the road surface disease detection results into a data attribute column, and uses GIS technology to import the underground radar map, road surface image and data attribute list into the road map to generate a road assessment map.

10. A computer-readable storage medium, which is applicable to the road exploration and evaluation method based on multidimensional data according to any one of claims 1 to 8, characterized in that: include: at least one processor (6) and memory (7); The memory (7) is used to store computer-executable instructions; The at least one processor (6) is used to execute the computer-executable instructions stored in the memory (7) to implement the road exploration and evaluation method based on multidimensional data as described in any one of claims 1 to 8.