UAV-Based Detection Method, Electronic Device, and Storage Medium for Cracks in Highway Concrete Pavements

Abstract

The invention relates to a method, electronic device and storage medium for detecting cracks in concrete pavement of highways based on drones, and belongs to the technical field of concrete structure disease detection. The invention aims to solve the problem that the accuracy of existing highway crack inspection technology is low. The invention sets the flight route and altitude of the drone, collects data on cracks in the concrete pavement of highways, and labels the crack categories using the labelimg tool. The crack categories are transverse cracks, longitudinal cracks, cracks and block cracks. The labeled data on cracks in the concrete pavement of highways are divided into a training set, a validation set and a test set. The YOLOV5 network model is improved to construct a network model for detecting cracks in concrete pavement of highways. The training set, validation set and test set are input into the network model for detecting cracks in concrete pavement of highways to obtain a trained network model for detecting cracks in concrete pavement of highways. The drone sets the route and controls the flight to realize detection. The invention realizes real-time and accurate detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of concrete structure disease detection, and specifically relates to a method, electronic equipment and storage medium for detecting cracks in concrete pavement of a highway based on an unmanned aerial vehicle (UAV). Background Art

[0002] Highway maintenance is a vital daily task for ensuring road safety, and cracks are a major factor affecting road quality. Currently, highway inspection methods are primarily divided into manual and semi-automated inspections. Manual inspections involve on-site surveys by highway inspection technicians, who manually measure the severity of road cracks and record and calculate them manually. This method is inefficient, poses safety risks, is time-consuming and labor-intensive, and is susceptible to subjective judgment by personnel, lacking standardized standards. Semi-automated inspections involve installing an automated detection system on inspection vehicles, which transmits real-time road footage to an on-board server and uses image algorithms to identify road defects. However, this method cannot overcome issues such as fallen trees and damaged road guardrails that affect normal traffic flow. Furthermore, the on-board inspection system, installed on the vehicle, has a limited field of view, limiting inspection efficiency. Furthermore, due to the viewing angle, it cannot accurately identify the severity of cracks. Summary of the Invention

[0003] The problem to be solved by the present invention is that the accuracy of existing highway crack inspection technology is not high. A method, electronic equipment and storage medium for detecting cracks in highway concrete pavement based on drone are proposed.

[0004] To achieve the above object, the present invention is implemented through the following technical solutions:

[0005] A method for detecting cracks in highway concrete pavement based on a drone comprises the following steps:

[0006] S1. Set the flight path and altitude of the drone to collect data on cracks in the concrete pavement of the highway.

[0007] S2. Using the labelimg tool, the highway concrete pavement crack data collected in step S1 is labeled with crack categories, including transverse cracks, longitudinal cracks, cracks, and block cracks. The labeled highway concrete pavement crack data is divided into a training set, a validation set, and a test set.

[0008] S3. Improve the YOLOV5 network model to build a highway concrete pavement crack detection network model;

[0009] Furthermore, the specific implementation method of step S3 includes the following steps:

[0010] S3.1. Input Layer: In the input layer, each batch of training data is passed through the data loader and fed into the neural network. The data is first normalized and then the selected image is concatenated with three other random images using Mosaic.

[0011] S3.2 Backbone network: The swin transformer is used as the backbone network, which mainly consists of the W-MSA structure and the SW-MSA structure. The W-MSA structure is used for mutual learning between patches within the feature window, and the SW-MSA structure is used for feature learning between windows at each layer.

[0012] S3.3, Neck connection network: Use BiFPN repeated weighted bidirectional feature pyramid network for weighted fusion, the weighting method is as follows,

[0013]

[0014] Among them, I i is the input of the i-th level, w i is the weight value of the i-th level of the current learning level, i is the current learning level, ε is a decimal used to prevent the denominator from being 0, w j is the weight value of the jth level after learning, j is the level after learning, and O is the weighted fusion feature;

[0015] Described by the fourth-level feature fusion of BiFPN, the calculation formula is:

[0016]

[0017]

[0018] Among them, C represents the convolution operation for feature processing, R represents the upsampling or downsampling operation for matching feature layers of different resolutions, is the intermediate feature at level 4 on the top-down path, is the intermediate feature at level 4 on the top-down path, Represents the output features of level 3 and level 4, P4 in 、P5 in are the input features of level 4 and level 5, w i ' is the intermediate weight on the top-down path;

[0019] S3.4. Final detection part: Apply anchor boxes to the feature map to make grid predictions on the feature map, set the NMS confidence parameters, retain the detected objects with confidence levels above the set threshold, and generate a final output vector with class probabilities. Heads of different scales are used to detect large, medium, and small objects respectively.

[0020] S4. Input the training set, validation set, and test set obtained in step S2 into the highway concrete pavement crack detection network model constructed in step S3 to obtain a trained highway concrete pavement crack detection network model;

[0021] S5. The drone sets a route or is remotely controlled to fly via a remote controller to inspect the highway. The images of highway concrete pavement cracks taken by the drone are transmitted to the highway concrete pavement crack detection network model trained in step S4 for detection.

[0022] Furthermore, the flight altitude of the UAV in step S1 is calculated based on the camera imaging principle and the geometric similarity principle, and the calculation formula is:

[0023]

[0024] Among them, H is the flight altitude of the UAV, Fr is the real focal length of the camera sensor, Dw is the real width of the ground covered by the image, and Sw is the real sensor width.

[0025] Furthermore, in step S2, the annotated highway concrete pavement crack data is divided into a training set, a validation set, and a test set according to a ratio of 8:1:1.

[0026] Furthermore, the specific implementation method of step S4 includes the following steps:

[0027] S4.1, the image passes through the first CBS module, which consists of Conv+BatchNorm+SiLU, with the number of channels set to 16 and the output size of the 0th layer feature map being 320;

[0028] S4.2. The layer 0 feature map obtained in step S4.1 is subjected to deep downsampling feature extraction by the CBS, C3, and C3STR modules to obtain a deep feature map;

[0029] S4.3. The deep feature map extracted in step S4.2 is pooled and fused through the SPPF module to obtain the backbone network output feature map;

[0030] S4.4. Upsampling the backbone network output feature map obtained in step S4.3, and performing feature fusion on the feature maps of different layers according to weights. For large objects that are less likely to lose information during downsampling, the prediction is made by fusion of the feature maps from the 10th and 21st layers. For small objects that experience pixel information loss as the depth of the convolutional layer increases, the prediction is made by fusion of the feature maps from the 4th and 15th layers, to obtain a feature map after feature fusion.

[0031] S4.5. Apply the anchor box to the feature map after feature fusion obtained in step S4.4 to make grid predictions on the feature map after feature fusion. Compare the prediction results with the label data to obtain the loss function. The calculation formula of the loss function LOSS is:

[0032] LOSS = classification loss cls_loss + positioning loss box_loss + confidence loss obj_loss

[0033] The weight parameters and thresholds are adjusted by backpropagation according to the loss function, the NMS confidence parameters are set, the detected targets with confidence higher than the set threshold are retained, and the final output vector with class probability is generated.

[0034] Furthermore, the specific implementation method of step S5 includes the following steps:

[0035] S5.1. The drone sets a route or is remotely controlled by a human to conduct highway inspections. Images of highway concrete pavement cracks are captured and transmitted to the trained highway concrete pavement crack detection network model.

[0036] S5.2. First, the highway concrete pavement crack image is normalized to unify the image input size and passed to the backbone network. Then, it is evenly divided into multiple windows through the Swin transformer network. The image features are fully learned between windows in the same layer and between windows in different layers.

[0037] S5.3. The network features extracted by the backbone are transferred to the neck BiFPN network for multiple multi-level weighted feature fusion, which combines strong semantic information with strong position information to classify and locate the detected target objects.

[0038] S5.4. The drone returns to the landing point according to the planned route.

[0039] The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps of the method for detecting cracks in concrete pavement of a highway based on a drone are implemented.

[0040] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for detecting cracks in concrete pavement of a highway based on a drone.

[0041] The beneficial effects of the present invention are:

[0042] The present invention discloses a method for detecting cracks in concrete pavement of highways based on drones, and a YOLOV5 neural network model: YOLOV5 is a one-stage target detection model with high speed and accuracy in neural network algorithms, comprising an input layer, a hidden layer, and an output layer. The network training process is a "forward propagation + reverse feedback" process. During the feedback process, the network parameters are iteratively updated according to the loss function to minimize the distance between the network prediction and the sample label.

[0043] The present invention describes a method for detecting cracks in highway concrete pavements using drones. This method aims to scientifically assess the technical condition and service level of highways and promote the scientific, standardized, and institutionalized development of highway technical condition assessment standards. The assessment criteria for crack severity based on crack width are shown in Table 1.

[0044] Table 1 Pavement crack assessment marking

[0045]

[0046] The method for detecting cracks in highway concrete pavement based on an unmanned aerial vehicle (UAV) described in the present invention is targeted at detecting cracks with a degree of damage of light or above.

[0047] The present invention discloses a method for detecting cracks in concrete pavement on highways based on drones. The drones replace manual inspections or inspection vehicles for highway inspections, saving manpower and material resources, reducing daily inspection costs, and increasing inspection speed. The YOLOV5 model itself has the outstanding advantages of a streamlined structure, fast detection speed, and high accuracy. The YOLOV5 neural network model is optimized for the small crack characteristics of highways, thereby improving the accuracy and speed of crack detection by the model. Inspection personnel can view highway pavement cracks and other conditions in real time.

[0048] The drone-based highway concrete pavement crack detection method described in the present invention proposes a new and advanced fully automatic detection method, which combines drones with the optimized neural network algorithm YOLOV5 to achieve real-time and accurate detection, overcoming the shortcomings of previous methods and greatly improving the quality, efficiency and cost of inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1This is a process flow chart of a method for detecting cracks in highway concrete pavement based on a drone according to the present invention;

[0050] Figure 2 This is a Neck structure diagram composed of BiFPN in the highway concrete pavement crack detection method based on drones described in the present invention;

[0051] Figure 3 This is a schematic diagram of the principle of calculating the flight altitude of a drone according to actual scene requirements in the drone-based highway concrete pavement crack detection method described in the present invention. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present invention and are not intended to limit the present invention. That is, the specific embodiments described herein are only some embodiments of the present invention, not all embodiments. Generally, the components of the specific embodiments of the present invention described and illustrated in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0053] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but is merely representative of selected specific embodiments of the present invention. All other specific embodiments obtained by those skilled in the art based on the specific embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0054] To further understand the content, features and effects of the present invention, the following specific embodiments are given as examples and described in detail with reference to the accompanying drawings: Specific implementation method one:

[0056] A method for detecting cracks in highway concrete pavement based on a drone comprises the following steps:

[0057] S1. Set the flight path and altitude of the drone to collect data on cracks in the concrete pavement of the highway.

[0058] Furthermore, the flight altitude of the UAV in step S1 is calculated based on the camera imaging principle and the geometric similarity principle, and the calculation formula is:

[0059]

[0060] Where H is the flight altitude of the UAV, Fr is the real focal length of the camera sensor, Dw is the real width of the ground covered by the image, and Sw is the real sensor width;

[0061] Furthermore, the experimental environment is a highway in Northeast China, consisting of three lanes and one emergency lane, with a total width of 14.6 meters. The reference flight altitude is calculated to be 9.73 meters based on the DJI Mavic 3 camera sensor parameters and the experimental road width;

[0062] Taking the DJI Mavic 3 as an example, the parameters of factors affecting implementation are shown in Table 2:

[0063] Table 2 UAV parameters

[0064]

[0065]

[0066] Data collection for cracks in highway concrete pavements is based on the collection of crack images in real-world scenarios. This requires a full range of crack types and a rich collection of scenarios, covering as many possible scenarios as possible during highway inspections, such as sunny, cloudy, and shadowed roads.

[0067] S2. Using the labelimg tool, the highway concrete pavement crack data collected in step S1 is labeled with crack categories, including transverse cracks, longitudinal cracks, cracks, and block cracks. The labeled highway concrete pavement crack data is divided into a training set, a validation set, and a test set.

[0068] Furthermore, in step S2, the annotated highway concrete pavement crack data is divided into a training set, a validation set, and a test set according to a ratio of 8:1:1;

[0069] S3. Improve the YOLOV5 network model to build a highway concrete pavement crack detection network model;

[0070] Furthermore, the specific implementation method of step S3 includes the following steps:

[0071] S3.1. Input Layer: In the input layer, each batch of training data is passed through the data loader and fed into the neural network. The data is first normalized and then the selected image is concatenated with three other random images using Mosaic.

[0072] S3.2 Backbone network: The swin transformer is used as the backbone network, which mainly consists of the W-MSA structure and the SW-MSA structure. The W-MSA structure is used for mutual learning between patches within the feature window, and the SW-MSA structure is used for feature learning between windows at each layer.

[0073] S3.3, Neck connection network: Use BiFPN repeated weighted bidirectional feature pyramid network for weighted fusion, the weighting method is as follows,

[0074]

[0075] Among them, I i is the input of the i-th level, w i is the weight value of the i-th level of the current learning level, i is the current learning level, ε is a decimal used to prevent the denominator from being 0, w j is the weight value of the jth level after learning, j is the level after learning, and O is the weighted fusion feature;

[0076] Described by the fourth-level feature fusion of BiFPN, the calculation formula is:

[0077]

[0078]

[0079] Among them, C represents the convolution operation for feature processing, R represents the upsampling or downsampling operation for matching feature layers of different resolutions, is the intermediate feature at level 4 on the top-down path, is the intermediate feature at level 4 on the top-down path, Represent the output features of level 3 and level 4 respectively, are the input features of level 4 and level 5, w i ' is the intermediate weight on the top-down path;

[0080] S3.4. Final detection part: Apply anchor boxes to the feature map to make grid predictions on the feature map, set the NMS confidence parameters, retain the detected objects with confidence levels above the set threshold, and generate a final output vector with class probabilities. Heads of different scales are used to detect large, medium, and small objects respectively.

[0081] S4. Input the training set, validation set, and test set obtained in step S2 into the highway concrete pavement crack detection network model constructed in step S3 to obtain a trained highway concrete pavement crack detection network model;

[0082] Furthermore, the network forward propagates: the data set is input into the network, image preprocessed by the input layer and then enters the hidden layer, and then enters the output layer through multi-layer weighted calculation to obtain the prediction result;

[0083] Furthermore, the specific implementation method of step S4 includes the following steps:

[0084] S4.1, the image passes through the first CBS module, which consists of Conv+BatchNorm+SiLU, with the number of channels set to 16 and the output size of the 0th layer feature map being 320;

[0085] S4.2. The layer 0 feature map obtained in step S4.1 is subjected to deep downsampling feature extraction by the CBS, C3, and C3STR modules to obtain a deep feature map;

[0086] S4.3. The deep feature map extracted in step S4.2 is pooled and fused through the SPPF module to obtain the backbone network output feature map;

[0087] S4.4. Upsampling the backbone network output feature map obtained in step S4.3, and performing feature fusion on the feature maps of different layers according to weights. For large objects that are less likely to lose information during downsampling, the prediction is made by fusion of the feature maps from the 10th and 21st layers. For small objects that experience pixel information loss as the depth of the convolutional layer increases, the prediction is made by fusion of the feature maps from the 4th and 15th layers, to obtain a feature map after feature fusion.

[0088] S4.5. Apply the anchor box to the feature map after feature fusion obtained in step S4.4 to make grid predictions on the feature map after feature fusion. Compare the prediction results with the label data to obtain the loss function. The calculation formula of the loss function LOSS is:

[0089] LOSS = classification loss cls_loss + positioning loss box_loss + confidence loss obj_loss

[0090] According to the loss function, the weight parameters and thresholds are adjusted through backpropagation, the NMS confidence parameters are set, the detected targets with confidence levels higher than the set threshold are retained, and the final output vector with class probabilities is generated;

[0091] S5. The drone sets a route or is remotely controlled to fly via a remote controller to inspect the highway. The images of highway concrete pavement cracks taken by the drone are transmitted to the highway concrete pavement crack detection network model trained in step S4 for detection.

[0092] Furthermore, the specific implementation method of step S5 includes the following steps:

[0093] S5.1. The drone sets a route or is remotely controlled by a human to conduct highway inspections. Images of highway concrete pavement cracks are captured and transmitted to the trained highway concrete pavement crack detection network model.

[0094] S5.2. First, the highway concrete pavement crack image is normalized to unify the image input size and passed to the backbone network. Then, it is evenly divided into multiple windows through the Swin transformer network. The image features are fully learned between windows in the same layer and between windows in different layers.

[0095] S5.3. The network features extracted by the backbone are transferred to the neck BiFPN network for multiple multi-level weighted feature fusion, which combines strong semantic information with strong position information to classify and locate the detected target objects.

[0096] S5.4. The drone returns to the landing point according to the planned route.

[0097] This embodiment describes a method for detecting cracks in highway concrete pavement using a drone. The drone mainly includes: an aircraft, a camera, a gimbal, a sensing system, an image transmission system, a battery, a remote control, etc., and proposes the following requirements:

[0098] 1. Has strong endurance;

[0099] 2. It can be used in most natural environments and has a certain wind resistance;

[0100] 3. Long-distance image signal transmission is stable;

[0101] 4. The camera has high resolution. To maximize the image field of view during flight, improve detection efficiency, and avoid near-ground obstacles for safe flight, the drone needs to increase its flight altitude. This places certain demands on the clarity of images collected from highway concrete cracks.

[0102] 5. Automatic obstacle avoidance: In natural environments, there are green areas and traffic signs on the sides of highways, and drones need to be able to avoid obstacles. Specific implementation method 2:

[0104] The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps of a method for detecting cracks in a highway concrete pavement based on an unmanned aerial vehicle (UAV) are implemented.

[0105] The computer device of the present invention may include a processor and memory, such as a single-chip microcomputer including a central processing unit. Furthermore, the processor is configured to execute a computer program stored in the memory to implement the steps of the aforementioned CREO software-based recommendation method for modifying relationship-driven recommendation data.

[0106] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0107] The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function or an image playback function); and the data storage area may store data generated based on the use of the mobile phone (such as audio data, a phone book, etc.). Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Specific implementation method three:

[0109] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for detecting cracks in concrete pavement of a highway based on a drone.

[0110] The computer-readable storage medium of the present invention can be any form of storage medium that can be read by a processor of a computer device, including but not limited to non-volatile memory, volatile memory, ferroelectric memory, etc. The computer-readable storage medium stores a computer program. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the above-mentioned modeling method of modifiable relationship-driven modeling data based on CREO software can be implemented.

[0111] The computer program includes computer program code, which may be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0112] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0113] Although the present application has been described above with reference to specific embodiments, various modifications may be made thereto and components may be substituted with equivalents without departing from the scope of the present application. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of these combinations in this specification is solely for the sake of space and resource conservation. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions within the scope of the claims.

Claims

1. A method for detecting cracks in highway concrete pavement based on drones, characterized by: The steps include: S1. Set the flight path and altitude of the drone to collect data on cracks in the concrete pavement of the highway. S2. Using the labelimg tool, the highway concrete pavement crack data collected in step S1 is labeled with crack categories, including transverse cracks, longitudinal cracks, cracks, and block cracks. The labeled highway concrete pavement crack data is divided into a training set, a validation set, and a test set. S3. Improve the YOLOV5 network model to build a highway concrete pavement crack detection network model; The specific implementation method of step S3 includes the following steps: S3.

1. Input Layer: In the input layer, each batch of training data is passed through the data loader and fed into the neural network. The data is first normalized and then the selected image is concatenated with three other random images using Mosaic. S3.2 Backbone network: The swin transformer is used as the backbone network, which mainly consists of the W-MSA structure and the SW-MSA structure. The W-MSA structure is used for mutual learning between patches within the feature window, and the SW-MSA structure is used for feature learning between windows at each layer. S3.3, Neck connection network: Use BiFPN repeated weighted bidirectional feature pyramid network for weighted fusion, the weighting method is as follows, Among them, I i is the input of the i-th level, w i is the weight value of the i-th level of the current learning level, i is the current learning level, ε is a decimal used to prevent the denominator from being 0, w j is the weight value of the jth level after learning, j is the level after learning, and O is the weighted fusion feature; Described by the fourth-level feature fusion of BiFPN, the calculation formula is: Among them, C represents the convolution operation for feature processing, R represents the upsampling or downsampling operation for matching feature layers of different resolutions, and P4 td is the intermediate feature at level 4 on the top-down path, P5 td is the intermediate feature at level 4 on the top-down path, P3 out 、P4 out Represents the output features P4 of level 3 and level 4 respectively in 、P5 in are the input features of level 4 and level 5, w i ' is the intermediate weight on the top-down path; S3.

4. Final detection part: Apply anchor boxes to the feature map to make grid predictions on the feature map, set the NMS confidence parameters, retain the detected objects with confidence levels above the set threshold, and generate a final output vector with class probabilities. Heads of different scales are used to detect large, medium, and small objects respectively. S4. Input the training set, validation set, and test set obtained in step S2 into the highway concrete pavement crack detection network model constructed in step S3 to obtain a trained highway concrete pavement crack detection network model; S5. The drone sets a route or is remotely controlled to fly via a remote controller to inspect the highway. The images of highway concrete pavement cracks taken by the drone are transmitted to the highway concrete pavement crack detection network model trained in step S4 for detection.

2. The method for detecting cracks in a highway concrete pavement based on an unmanned aerial vehicle according to claim 1, characterized in that: In step S1, the flight altitude of the UAV is calculated based on the camera imaging principle and the principle of geometric similarity. The calculation formula is: Among them, H is the flight altitude of the UAV, Fr is the real focal length of the camera sensor, Dw is the real width of the ground covered by the image, and Sw is the real sensor width.

3. The method for detecting cracks in a highway concrete pavement based on an unmanned aerial vehicle according to claim 1 or 2, characterized in that: In step S2, the annotated highway concrete pavement crack data is divided into a training set, a validation set, and a test set in a ratio of 8:1:

1.

4. The method for detecting cracks in a highway concrete pavement based on an unmanned aerial vehicle according to claim 3, characterized in that: The specific implementation method of step S4 includes the following steps: S4.1, the image passes through the first CBS module, which consists of Conv+BatchNorm+SiLU, with the number of channels set to 16 and the output size of the 0th layer feature map being 320; S4.

2. The layer 0 feature map obtained in step S4.1 is subjected to deep downsampling feature extraction by the CBS, C3, and C3STR modules to obtain a deep feature map; S4.

3. The deep feature map extracted in step S4.2 is pooled and fused through the SPPF module to obtain the backbone network output feature map; S4.

4. Upsampling the backbone network output feature map obtained in step S4.3, and performing feature fusion on the feature maps of different layers according to weights. For large objects that are less likely to lose information during downsampling, the prediction is made by fusion of the feature maps from the 10th and 21st layers. For small objects that experience pixel information loss as the depth of the convolutional layer increases, the prediction is made by fusion of the feature maps from the 4th and 15th layers, to obtain a feature map after feature fusion. S4.

5. Apply the anchor box to the feature map after feature fusion obtained in step S4.4 to make grid predictions on the feature map after feature fusion. Compare the prediction results with the label data to obtain the loss function. The calculation formula of the loss function LOSS is: LOSS = classification loss cls_loss + positioning loss box_loss + confidence loss obj_loss The weight parameters and thresholds are adjusted by backpropagation according to the loss function, the NMS confidence parameters are set, the detected targets with confidence higher than the set threshold are retained, and the final output vector with class probability is generated.

5. The method for detecting cracks in highway concrete pavement based on an unmanned aerial vehicle according to claim 4, characterized in that: The specific implementation method of step S5 includes the following steps: S5.

1. The drone sets a route or is remotely controlled by a human to conduct highway inspections. Images of highway concrete pavement cracks are captured and transmitted to the trained highway concrete pavement crack detection network model. S5.

2. First, the highway concrete pavement crack image is normalized to unify the image input size and passed to the backbone network. After that, it is evenly divided into multiple windows by the swintransformer network. The image features are fully learned between windows in the same layer and between windows in different layers. S5.

3. The network features extracted by the backbone are transferred to the neck BiFPN network for multiple multi-level weighted feature fusion, which combines strong semantic information with strong position information to classify and locate the detected target objects. S5.

4. The drone returns to the landing point according to the planned route.

6. An electronic device, characterized in that The method comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for detecting cracks in concrete pavement of a highway based on a drone are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting cracks in concrete pavement of a highway based on a drone as described in any one of claims 1 to 5 is implemented.

Images