Road crack detection system and method thereof
Through mobile range data collection and preprocessing technology, a convolutional neural network model is established, combined with accelerometer and coordinate components, the omission-free detection of the highway crack detection system is achieved, and the problems of small detection range and low efficiency in the existing technology are solved. Improves detection efficiency and accuracy.
Patent Information
- Application Number
- CN202410958670.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing highway crack detection system uses a mobile vehicle to drive the detection head for inspection. The detection range is small and can only be performed in a linear manner, resulting in omissions and low efficiency.
Moving range data collection and preprocessing technology are used to establish a mobile range model, identify the road detection range through convolutional neural network, and coordinate motion is realized through accelerometers and coordinate components to ensure no missed detection.
It improves the accuracy of positioning detection within the detection range, avoids equipment delays and omissions, and enhances detection efficiency.
Smart Images

Figure CN118914530A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of highway crack detection, and more specifically, to a highway crack detection system and method thereof. Background Art
[0002] Highways are traffic signs in cities. After long-term use, cracks may appear on the road surface due to time or the passage of vehicles of different weights. Highway crack detection is necessary, especially for timely detection and feedback of road surfaces that are about to crack or have internal problems.
[0003] The document with the prior art publication number CN109763409B provides a highway crack detection system. The device is connected to the main controller and a traffic department management module. The traffic department management module stores a traffic road map of the highway within the management scope and divides the traffic road map into different areas. Different areas correspond to relevant information of different managers. The relevant information includes name, mobile phone number and ID card information. The system is arranged on a detection vehicle for highway detection. The acquisition end includes, which is connected to the main controller for communication. The image acquisition module is used to collect highway road image information. The image acquisition module is connected to the image acquisition module. The judgment module includes a storage module and a comparison module. The storage module stores normal highway road image information, image information of highway cracks with different cracking degrees, and image information of highway cracks with different cracking degrees. Grade classification data information; the comparison module compares the image information collected by the image acquisition module with the image information in the storage module and classifies the highway pavement with cracks, which is connected to the traffic department management module; the timing module starts counting down the time limit for repairing pavement cracks when the grade classification of pavement cracks is marked on the traffic pavement map, and is connected to the acquisition end and the timing module respectively; the feedback module collects the image of the repaired pavement according to the acquisition end and compares it with the image in the storage module through the comparison module; if the pavement is repaired within the specified time and meets the requirements, the timing module cancels the mark on the traffic pavement map; if the pavement repair is not completed within the specified time or does not meet the requirements, the timing module starts overtime timing from the end of the countdown until the pavement meets the requirements and the timing module ends.
[0004] Although the above-mentioned prior art solution can achieve the relevant beneficial effects through the structure of the prior art, it still has the following defects: the device drives the detection head to perform detection through a mobile vehicle, and only through the forward and backward movement of the mobile vehicle, not only the detection range is small, but also only linear detection can be performed, and there is a gap in the detection when the detection is acquired, which may lead to omissions.
[0005] Regarding the above-mentioned related technologies, the inventors believe that during the detection process, not only is it necessary for the detection head to cover a certain range area for detection to improve the detection efficiency, but it is also necessary to detect each location within the detection range to avoid device delays or other omissions.
[0006] In view of this, we propose a highway crack detection system. Summary of the invention
[0007] 1. Technical problems to be solved
[0008] The purpose of the present application is to provide a method for a highway crack detection system, which solves the technical problem that the device in the above-mentioned background technology uses a mobile vehicle to drive the detection head for detection, and only through the forward and backward movement of the mobile vehicle, not only the detection range is small, but also only linear detection can be performed, and there is a gap in the detection when the detection is obtained, thereby causing omissions, thereby achieving a technical effect.
[0009] 2. Technical solution
[0010] The technical solution of the present application provides a highway crack detection system, comprising
[0011] A module for collecting moving range data collects the range data of the moving points of the detection head;
[0012] The moving range data preprocessing module resizes, removes noise, and enhances contrast of the moving range data to ensure that the model can better learn features;
[0013] The moving range data feature extraction module converts the moving range data into raw data so that the model can learn;
[0014] Establishing a moving range model, using the moving range data to establish a moving range model;
[0015] A moving range model training module uses the prepared data set to train the moving range model so that the moving range model can accurately identify the range of road detection;
[0016] Moving range model evaluation and tuning, using the validation set to evaluate the trained moving range model to measure its performance on unseen data;
[0017] Mobile range model deployment monitoring, deploying the trained model to the actual environment for testing;
[0018] The execution module allows the detection head to perform coordinate movement based on the prediction results, thereby performing a complete detection.
[0019] By adopting the above technical solution and setting the mobile range model, each location within the detection range can be detected during the detection process to avoid device delays or other omissions, and improve the efficiency of precise mobile positioning through continuous learning.
[0020] As an optional solution of the technical solution of this application document, the module for collecting movement range data uses an accelerometer to acquire data, and the accelerometer is arranged on the outer wall of the detection head;
[0021] The accelerometer is used to measure the linear acceleration of an object. Based on Newton's second law, the acceleration of the object is calculated by measuring the force on a unit mass point. When the detection head moves, acceleration is generated. By monitoring the change in acceleration, the motion state of the device can be detected, such as acceleration, deceleration, and stillness.
[0022] As an optional solution of the technical solution of the present application document, the moving range data preprocessing module adjusts the image to a uniform size to ensure that the images input into the moving range model have the same dimension;
[0023] The noise in the image may interfere with the learning process of the moving range model. Filtering technology is used to remove the noise and increase the model's attention to useful features.
[0024] Enhancing the contrast of the image makes it easier for the moving range model to distinguish details and features in the image, which is achieved through methods such as histogram equalization.
[0025] As an optional solution of the technical solution of this application document, the moving range data feature extraction module uses edge detection image processing technology to extract manual features of the image;
[0026] Convert the text into count vectors of the words in the vocabulary. Each document can be represented as a vector containing the number of times each word appears in the document.
[0027] Extract statistical information of time series, including mean, standard deviation, maximum value, minimum value, etc.;
[0028] Use sliding window techniques to extract statistics within a window to account for local patterns in time;
[0029] Convert time series to the frequency domain to capture periodic patterns;
[0030] For numerical data, statistical features include mean, standard deviation, median, etc. High powers of features are added to take into account nonlinear relationships between variables and to create new features, such as the product or quotient of two variables.
[0031] As an optional solution of the technical solution of the present application document, the moving range model is established, and a convolutional neural network is selected to process image data. Through convolution layers and pooling layers, CNN can automatically learn local and global features in the image.
[0032] As an optional solution of the technical solution of the present application document, the moving range model training module uses a training set to train the model. In each training iteration, image data and corresponding labels are provided to the model, and the gradient is calculated according to the loss function;
[0033] The model weights are updated through the optimization algorithm, and the moving range model is gradually adjusted to improve the recognition ability of the current range detection of the road.
[0034] As an optional solution of the technical solution of the present application document, the moving range model evaluation and tuning uses architecture modification to improve performance, tries different CNN structures, and if it is a deeper network, adds residual connections to help solve the gradient vanishing problem and improve the training effect;
[0035] Adding batch normalization layers helps speed up moving range model training and improves the robustness of the moving range model to hyperparameter choices.
[0036] As an optional solution of the technical solution of this application document, the execution module includes
[0037] Inspection vehicle;
[0038] A connecting frame, wherein the connecting frame is slidably arranged on the bottom surface of the inspection vehicle through a moving assembly;
[0039] A detection and positioning block is used for coordinate positioning through a coordinate component.
[0040] The present application also discloses a method for using the aforementioned highway crack detection system, comprising the following steps:
[0041] S1. First, collect the range data of the moving points of the detection head, use the accelerometer to measure the linear acceleration of the object, and calculate the acceleration of the object by measuring the force on the unit mass point based on Newton's second law. When the detection head moves, acceleration will be generated. By monitoring the change of acceleration, the movement state of the device can be detected, such as acceleration, deceleration, and stillness;
[0042] S2. At this time, the preprocessing of resizing, denoising, and contrast enhancement of the moving range data is to adjust the image to a uniform size to ensure that the images input to the moving range model have the same dimension. The noise in the image may interfere with the learning process of the moving range model. The noise is removed by filtering technology to improve the model's attention to useful features. The contrast of the image is enhanced to make it easier for the moving range model to distinguish details and features in the image. This is achieved through methods such as histogram equalization to ensure that the model can better learn features.
[0043] S3. At this time, extract the features of the moving range data. Use edge detection image processing technology to extract manual features of the image. Convert the text into a count vector of words in the vocabulary. Each document can be represented as a vector containing the number of times each word appears in the document. Extract the statistical information of the time series, such as mean, standard deviation, maximum, minimum, etc. Use sliding window technology to extract statistical information within the window to consider local patterns in time. Convert the time series to the frequency domain to capture periodic patterns. For numerical data, statistical features include mean, standard deviation, median, etc. Add high powers of features to consider nonlinear relationships between variables and create new features, such as the product or quotient of two variables.
[0044] S4. At this point, a moving range model is established and a convolutional neural network is selected to process image data. Through convolutional layers and pooling layers, CNN can automatically learn local and global features in the image.
[0045] S5. At this point, the model is trained using the training set. In each training iteration, the model is provided with image data and corresponding labels, and the gradient is calculated according to the loss function. The weight of the model is updated through the optimization algorithm, and the moving range model is gradually adjusted to improve the recognition ability of the current range detection of the road;
[0046] S6. At this point, use architecture modifications to improve performance and try different CNN structures. If it is a deeper network, add residual connections to help solve the gradient vanishing problem and improve training results. Adding batch normalization layers can help speed up the training of the moving range model and improve the robustness of the moving range model to the selection of hyperparameters.
[0047] S7. At this point, the trained model is deployed to the actual environment for detection, and the detection head is moved for detection based on the prediction.
[0048] 3. Beneficial effects
[0049] One or more technical solutions provided in the technical solution of this application have at least the following technical effects or advantages:
[0050] 1. This application can detect each location within the detection range during the detection process by setting up a mobile range model, avoiding device delays or other omissions, and improving the efficiency of precise mobile positioning through continuous learning.
[0051] 2. This application can allow the detection head to cover a certain range of areas for detection through the setting of coordinate components, thereby improving the efficiency of detection.
[0052] 4. Through the setting of the mobile component, the present application can move the detection head to the outside of the detection vehicle for detection, thereby further increasing the coverage range. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 A schematic diagram of the structure of a highway crack detection system disclosed in a preferred embodiment of the present application;
[0054] Figure 2 This is a schematic diagram of the overall structure of a highway crack detection system disclosed in a preferred embodiment of the present application;
[0055] Figure 3 A bottom view of the overall structure of a highway crack detection system disclosed in a preferred embodiment of the present application;
[0056] Figure 4 This is a structural disassembly diagram of a mobile component of a highway crack detection system disclosed in a preferred embodiment of the present application;
[0057] Figure 5 This is an enlarged view of the structure at location A of a highway crack detection system disclosed in a preferred embodiment of the present application;
[0058] Figure 6 An exploded diagram of the coordinate component structure of a highway crack detection system disclosed in a preferred embodiment of the present application;
[0059] Figure 7 A partial structural breakdown diagram of a coordinate component of a highway crack detection system disclosed in a preferred embodiment of the present application;
[0060] Figure 8 This is an enlarged view of the structure at position B of a highway crack detection system disclosed in a preferred embodiment of the present application;
[0061] Explanation of the numbers in the figure: 1. Inspection vehicle; 11. Push handle; 12. Storage slot; 2. Connecting frame; 21. Slider; 22. Slide slot; 23. Drive slot; 24. Cam; 25. Chassis; 26. First motor; 27. Guide shaft; 3. Inspection positioning block; 4. Coordinate assembly; 41. Moving slot; 42. Coordinate rod; 43. Limiting slot; 44. Synchronous block; 45. Sliding slot; 46. Rotating chamber; 47. Drive shaft; 48. Driving slot; 49. Pulley; 410. Mechanical box; 411. Second motor; 5. Inspection head. DETAILED DESCRIPTION
[0062] The present application is further described in detail below in conjunction with the accompanying drawings.
[0063] Reference Figure 1 , the embodiment of the present application provides a highway crack detection system, a moving range data collection module, which collects the range data of the moving points of the detection head 5;
[0064] The moving range data preprocessing module resizes, removes noise, and enhances contrast of the moving range data to ensure that the model can better learn features;
[0065] The moving range data feature extraction module converts the moving range data into raw data so that the model can learn;
[0066] Establishing a moving range model, using the moving range data to establish a moving range model;
[0067] A moving range model training module uses the prepared data set to train the moving range model so that the moving range model can accurately identify the range of road detection;
[0068] Moving range model evaluation and tuning, using the validation set to evaluate the trained moving range model to measure its performance on unseen data;
[0069] Mobile range model deployment monitoring, deploying the trained model to the actual environment for testing;
[0070] The execution module allows the detection head 5 to perform coordinate movement according to the prediction result, so as to perform a complete detection.
[0071] By setting up the mobile range model, each location within the detection range can be detected during the detection process to avoid device delays or other omissions, and improve the efficiency of precise mobile positioning through continuous learning.
[0072] Reference Figure 1 and Figure 2The embodiment of the present application provides a highway crack detection system, a module for collecting movement range data, and an accelerometer is used to acquire data. The accelerometer is arranged on the outer wall of the detection head 5;
[0073] The accelerometer is used to measure the linear acceleration of an object. Based on Newton's second law, the acceleration of the object is calculated by measuring the force on a unit mass point. When the detection head 5 moves, acceleration will be generated. By monitoring the change in acceleration, the motion state of the device can be detected, such as acceleration, deceleration, and stillness.
[0074] The moving range data preprocessing module resizes the images to a uniform size to ensure that the images input to the moving range model have the same dimensions;
[0075] The noise present in the image may interfere with the learning process of the moving range model. Filtering techniques such as Gaussian filtering can be used to remove the noise and increase the model's attention to useful features.
[0076] Enhancing the contrast of the image makes it easier for the moving range model to distinguish details and features in the image, which is achieved through methods such as histogram equalization.
[0077] The moving range data feature extraction module uses edge detection image processing technology to extract manual features of the image;
[0078] Convert the text into count vectors of the words in the vocabulary. Each document can be represented as a vector containing the number of times each word appears in the document.
[0079] Extract statistical information of time series, including mean, standard deviation, maximum value, minimum value, etc.;
[0080] Use sliding window techniques to extract statistics within a window to account for local patterns in time;
[0081] Convert time series to the frequency domain to capture periodic patterns;
[0082] For numerical data, statistical features include mean, standard deviation, median, etc. High powers of features are added to take into account nonlinear relationships between variables and to create new features, such as the product or quotient of two variables.
[0083] A moving range model is established and a convolutional neural network (CNN) is selected to process image data. Through convolutional layers and pooling layers, CNN can automatically learn local and global features in images.
[0084] The mobile range model training module trains the model using the training set. In each training iteration, the model is provided with image data and corresponding labels, and the gradient is calculated according to the loss function;
[0085] The model weights are updated through the optimization algorithm, and the moving range model is gradually adjusted to improve the recognition ability of the current range detection of the road.
[0086] Mobile range model evaluation and tuning, using architecture modifications to improve performance, trying different CNN structures, and if it is a deeper network, adding residual connection ResNet to help solve the gradient disappearance problem and improve training results;
[0087] Adding batch normalization layers helps speed up moving range model training and improves the robustness of the moving range model to hyperparameter choices.
[0088] Reference Figure 2 and Figure 3 The present application embodiment provides a highway crack detection system, comprising:
[0089] Inspection vehicle 1;
[0090] A connecting frame 2, the connecting frame 2 is slidably arranged on the bottom surface of the inspection vehicle 1 through a moving component;
[0091] The detection and positioning block 3 is coordinate-positioned by the coordinate component 4 .
[0092] A push handle 11 is fixedly arranged on the outside of the inspection vehicle 1 , and a storage slot 12 is provided on the surface of the inspection vehicle 1 .
[0093] Other necessary tools can be placed through the storage slot 12.
[0094] Reference Figure 4 and Figure 5 The embodiment of the present application provides a highway crack detection system, wherein the moving assembly includes a slider 21 fixedly arranged on the outer wall of the detection positioning block 3, wherein two sliders 21 are arranged in a pair structure, and the sliders 21 are slidably arranged inside a slide groove 22 opened on the outer wall of the detection vehicle 1;
[0095] A guide shaft 27 is rotatably arranged inside the slide groove 22, and the guide shaft 27 passes through the inside and outside of the slider 21 and is slidably connected thereto;
[0096] A driving groove 23 is formed on the outer wall of one of the guide shafts 27. The driving groove 23 is in a spiral structure. A cam 24 is slidably disposed inside the driving groove 23. The cam 24 is fixedly disposed on the inner wall of the slider 21.
[0097] A chassis 25 is fixedly disposed on the outer wall of the inspection vehicle 1 , a first motor 26 is disposed inside the chassis 25 , and an output shaft of the first motor 26 is coaxially and fixedly connected with a guide shaft 27 .
[0098] The guide shaft 27 is driven to rotate by the rotation of the output shaft of the first motor 26, and the driving groove 23 on the guide shaft 27 is limited by the cam 24, so that the slider 21 can move, and the entire detection positioning block 3 can move to expand the range. Through the setting of the moving component, the detection head can be moved to the outside of the detection vehicle for detection, thereby further increasing the coverage range.
[0099] Reference Figure 6 , Figure 7 and Figure 8 The embodiment of the present application provides a highway crack detection system, wherein the coordinate assembly 4 includes a moving groove 41 provided on the outer wall of the connecting frame 2, a coordinate rod 42 is slidably provided inside the moving groove 41, and one side of the coordinate rod 42 is slidably provided inside a limiting groove 43 provided on the inner wall of the moving groove 41;
[0100] Two coordinate rods 42 are arranged in a vertical structure, and the two coordinate rods 42 are connected by a synchronization block 44. Two sliding grooves 45 are arranged on the outer wall of the synchronization block 44. The sliding grooves 45 are arranged in an up-and-down structure and are vertical. The sliding grooves 45 are slidably connected to the coordinate rods 42. The outer wall of the coordinate rods 42 is fixedly provided with a detection head 5;
[0101] A rotating chamber 46 is provided inside the connecting frame 2, a driving shaft 47 is rotatably provided inside the rotating chamber 46, a driving groove 48 is provided on the outer wall of the driving shaft 47, the driving groove 48 is provided in an annular curved structure, a pulley 49 is slidably provided inside the driving groove 48, and the pulley 49 is fixedly provided at one end of the coordinate rod 42;
[0102] A mechanical box 410 is fixedly disposed on the outer wall of the connecting frame 2 , and a second motor 411 is disposed inside the mechanical box 410 . The output shaft of the second motor 411 is coaxially fixedly connected to the driving shaft 47 .
[0103] The output shaft of the second motor 411 drives the driving shaft 47 to rotate, and the driving groove 48 on the driving shaft 47 prompts the pulley 49 to drive the coordinate rod 42 to move, thereby driving the synchronization block 44 to move, allowing the detection head 5 to be positioned in one direction. At the same time, the vertical coordinate rod 42 on the other side will drive the synchronization block 44 to move in another vertical direction, so that the detection head 5 can form a surface detection range. Through the setting of the coordinate component, the detection head can cover a certain range area for detection, thereby improving the detection efficiency.
[0104] The present application embodiment provides a highway crack detection system, comprising the following steps:
[0105] S1. First, collect the range data of the moving points of the detection head 5, use the accelerometer to measure the linear acceleration of the object, and calculate the acceleration of the object by measuring the force on the unit mass point based on Newton's second law. When the detection head 5 moves, acceleration will be generated. By monitoring the change of acceleration, the movement state of the device can be detected, such as acceleration, deceleration, and stillness;
[0106] S2. At this time, the preprocessing of resizing, denoising, and contrast enhancement of the moving range data is to adjust the image to a uniform size to ensure that the images input to the moving range model have the same dimension. The noise in the image may interfere with the learning process of the moving range model. The noise is removed by filtering techniques such as Gaussian filtering to increase the model's attention to useful features. The contrast of the image is enhanced to make it easier for the moving range model to distinguish details and features in the image. This is achieved through methods such as histogram equalization to ensure that the model can better learn features.
[0107] S3. At this time, extract the features of the moving range data. Use edge detection image processing technology to extract manual features of the image. Convert the text into a count vector of words in the vocabulary. Each document can be represented as a vector containing the number of times each word appears in the document. Extract the statistical information of the time series, such as mean, standard deviation, maximum, minimum, etc. Use sliding window technology to extract statistical information within the window to consider local patterns in time. Convert the time series to the frequency domain to capture periodic patterns. For numerical data, statistical features include mean, standard deviation, median, etc. Add high powers of features to consider nonlinear relationships between variables and create new features, such as the product or quotient of two variables.
[0108] S4. At this point, a moving range model is established and a convolutional neural network (CNN) is selected to process image data. Through convolutional layers and pooling layers, CNN can automatically learn local and global features in the image.
[0109] S5. At this point, the model is trained using the training set. In each training iteration, the model is provided with image data and corresponding labels, and the gradient is calculated according to the loss function. The weight of the model is updated through the optimization algorithm, and the moving range model is gradually adjusted to improve the recognition ability of the current range detection of the road;
[0110] S6. At this point, use architecture modifications to improve performance and try different CNN structures. If it is a deeper network, add residual connection ResNet to help solve the gradient vanishing problem and improve training results. Adding batch normalization layers helps speed up the training of the moving range model and improve the robustness of the moving range model to the selection of hyperparameters.
[0111] S7. At this time, the trained model is deployed to the actual environment for detection, and the detection head 5 is moved for detection according to the prediction.
[0112] When the highway crack detection system is needed, first, the range data of the moving points of the detection head 5 is collected, and the linear acceleration of the object is measured by an accelerometer. Based on Newton's second law, the acceleration of the object is calculated by measuring the force on the unit mass point. When the detection head 5 moves, acceleration will be generated. By monitoring the change in acceleration, the movement state of the equipment can be detected, such as acceleration, deceleration, and stillness. At this time, the preprocessing of resizing, denoising, and contrast enhancement of the moving range data is to adjust the image to a uniform size to ensure that the images input into the moving range model have the same dimension. The noise present in the image may interfere with the learning process of the moving range model. Filtering techniques such as Gaussian filtering are used to remove noise, improve the model's attention to useful features, and enhance the image contrast so that the moving range model can more easily distinguish details and features in the image. This is achieved through methods such as histogram equalization to ensure that the model can better learn features. At this time, the moving range data features are extracted, and edge detection image processing technology is used to extract manual features of the image and convert text into Count vectors of words in the vocabulary. Each document can be represented as a vector containing the number of times each word appears in the document. Extract statistical information of the time series, such as mean, standard deviation, maximum, minimum, etc. Use sliding window technology to extract statistical information within the window to consider local patterns in time. Convert the time series to the frequency domain to capture periodic patterns. For numerical data, statistical features include mean, standard deviation, median, etc. Add high powers of features to consider nonlinear relationships between variables. Create new features, such as the product or quotient of two variables. Build a moving range model. Select convolutional neural network CNN to process image data. Through convolutional layers and pooling layers, CNN can automatically learn local and global features in images. Train the model using the training set.In each training iteration, the model is provided with image data and corresponding labels, and the gradient is calculated according to the loss function. The weight of the model is updated through the optimization algorithm, and the mobile range model is gradually adjusted to improve the recognition ability of the current range detection of the road. The architecture modification is used to improve the performance. Different CNN structures are tried. If it is a deeper network, adding residual connection ResNet helps solve the gradient disappearance problem and improves the training effect. Adding batch normalization layer helps to accelerate the training of the mobile range model and improve the robustness of the mobile range model to the selection of hyperparameters. The trained model is deployed to the actual environment for detection. According to the prediction, the detection head is moved 5 times for detection, so as to control the detection. The output shaft of the second motor 411 drives the driving shaft 47 to rotate, and the driving groove 48 on the driving shaft 47 prompts the pulley 49 to drive the coordinate rod 42 to move, thereby driving the synchronization block 44 to move, allowing the detection head 5 to be positioned in one direction, and at the same time, the vertical coordinate rod 42 on the other side will drive the synchronization block 44 to move in another vertical direction, so that the detection head 5 can form a detection range of a surface, and at the same time, the guide shaft 27 can be driven to rotate by the rotation of the output shaft of the first motor 26, and the driving groove 23 on the guide shaft 27 is limited by the cam 24, so that the slider 21 can move its position, thereby allowing the entire detection positioning block 3 to move its position to further expand its range.
Claims
1. A highway crack detection system, characterized in that: Include: A module for collecting moving range data collects the range data of the moving points of the detection head; The moving range data preprocessing module resizes, removes noise, and enhances contrast of the moving range data to ensure that the model can better learn features; The moving range data feature extraction module converts the moving range data into raw data so that the model can learn; Establishing a moving range model, using the moving range data to establish a moving range model; A moving range model training module uses the prepared data set to train the moving range model so that the moving range model can accurately identify the range of road detection; Moving range model evaluation and tuning, using the validation set to evaluate the trained moving range model to measure its performance on unseen data; Mobile range model deployment monitoring, deploying the trained model to the actual environment for testing; The execution module allows the detection head to perform coordinate movement based on the prediction results, thereby performing a complete detection.
2. The highway crack detection system according to claim 1, characterized in that: The module for collecting movement range data acquires data using an accelerometer, and the accelerometer is arranged on the outer wall of the detection head; An accelerometer is used to measure the linear acceleration of an object. Based on Newton's second law, the acceleration of an object is calculated by measuring the force on a unit mass point. When the detection head moves, acceleration is generated. By monitoring the changes in acceleration, the motion state of the device can be detected, such as acceleration, deceleration, and stillness.
3. The highway crack detection system according to claim 1, characterized in that: The moving range data preprocessing module adjusts the images to a uniform size to ensure that the images input into the moving range model have the same dimensions; The noise in the image may interfere with the learning process of the moving range model. Filtering technology is used to remove the noise and increase the model's attention to useful features. Enhancing the contrast of the image makes it easier for the moving range model to distinguish details and features in the image, which is achieved through methods such as histogram equalization.
4. The highway crack detection system according to claim 1, characterized in that: The moving range data feature extraction module uses edge detection image processing technology to extract manual features of the image; Convert the text into count vectors of the words in the vocabulary. Each document can be represented as a vector containing the number of times each word appears in the document. Extract statistical information of time series, including mean, standard deviation, maximum value, minimum value, etc.; Use sliding window techniques to extract statistics within a window to account for local patterns in time; Convert time series to the frequency domain to capture periodic patterns; For numerical data, statistical features include mean, standard deviation, median, etc. High powers of features are added to take into account nonlinear relationships between variables and to create new features, such as the product or quotient of two variables.
5. The highway crack detection system according to claim 1, characterized in that: The moving range model is established, and a convolutional neural network is selected to process image data. Through the convolution layer and the pooling layer, the CNN can automatically learn the local and global features in the image.
6. The highway crack detection system according to claim 1, characterized in that: The moving range model training module trains the model using the training set, provides image data and corresponding labels to the model in each training iteration, and calculates the gradient according to the loss function; The model weights are updated through the optimization algorithm, and the moving range model is gradually adjusted to improve the recognition ability of the current range detection of the road.
7. The highway crack detection system according to claim 1, characterized in that: The mobile range model evaluation and tuning uses architecture modifications to improve performance and try different CNN structures. If it is a deeper network, adding residual connections helps solve the gradient vanishing problem and improves training results. Adding batch normalization layers helps speed up moving range model training and improves the robustness of the moving range model to hyperparameter choices.
8. The highway crack detection system according to claim 1, characterized in that: The execution module includes Inspection vehicle; A connecting frame, wherein the connecting frame is slidably arranged on the bottom surface of the inspection vehicle through a moving assembly; A detection and positioning block is used for coordinate positioning through a coordinate component.
9. A method for using a highway crack detection system according to any one of claims 1 to 8, characterized in that: The following steps are involved: S1. First, collect the range data of the moving points of the detection head, use the accelerometer to measure the linear acceleration of the object, and calculate the acceleration of the object by measuring the force on the unit mass point based on Newton's second law. When the detection head moves, acceleration will be generated. By monitoring the change of acceleration, the movement state of the device can be detected, such as acceleration, deceleration, and stillness; S2. At this time, the preprocessing of resizing, denoising, and contrast enhancement of the moving range data is to adjust the image to a uniform size to ensure that the images input to the moving range model have the same dimension. The noise in the image may interfere with the learning process of the moving range model. The noise is removed by filtering technology to improve the model's attention to useful features. The contrast of the image is enhanced to make it easier for the moving range model to distinguish details and features in the image. This is achieved through methods such as histogram equalization to ensure that the model can better learn features. S3. At this time, extract the features of the moving range data. Use edge detection image processing technology to extract manual features of the image. Convert the text into a count vector of words in the vocabulary. Each document can be represented as a vector containing the number of times each word appears in the document. Extract the statistical information of the time series, such as mean, standard deviation, maximum, minimum, etc. Use sliding window technology to extract statistical information within the window to consider local patterns in time. Convert the time series to the frequency domain to capture periodic patterns. For numerical data, statistical features include mean, standard deviation, median, etc. Add high powers of features to consider nonlinear relationships between variables and create new features, such as the product or quotient of two variables. S4. At this point, a moving range model is established and a convolutional neural network is selected to process image data. Through convolutional layers and pooling layers, CNN can automatically learn local and global features in the image. S5. At this time, the model is trained using the training set. In each training iteration, the model is provided with image data and corresponding labels, and the gradient is calculated according to the loss function. The weight of the model is updated through the optimization algorithm, and the moving range model is gradually adjusted to improve the recognition ability of the current range detection of the road. S6. At this point, use architecture modifications to improve performance and try different CNN structures. If it is a deeper network, add residual connections to help solve the gradient vanishing problem and improve training results. Adding batch normalization layers can help speed up the training of the moving range model and improve the robustness of the moving range model to the selection of hyperparameters. S7. At this point, the trained model is deployed to the actual environment for detection, and the detection head is moved for detection based on the prediction.
Citation Information
Patent Citations
Highway Crack Detection System
CN109763409B
