Road inspection disease de-weighting method and device, electronic equipment and storage medium

Abstract

The invention is suitable for the technical field of smart cities, and provides a road inspection disease deduplication method, which comprises the following steps: collecting road disease data from different mobile inspection devices; performing space-time fusion processing on the road disease data through a preset space-time fusion model to obtain road disease fusion features; calculating an optimal transmission distance between the road disease fusion features through a preset entropy regularization algorithm, and generating a cost matrix; and based on the cost matrix, determining target road disease data, and performing de-duplication processing on the target road disease data. The method solves the problems that an existing disease duplicate removal method depends on static visual features of diseases, the essential similarity of the diseases under geometric deformation cannot be effectively captured, and the same disease is easily and repeatedly detected.

Description

Technical Field

[0001] This application belongs to the field of smart city technology, and in particular relates to a method, device, electronic device and storage medium for deduplication of road inspection defects. Background Technology

[0002] Road defects refer to various types of damage, deformation, and other defects that occur on roads. Common road defects include cracks, potholes, loosening, and subsidence. The presence of road defects not only shortens the service life of roads but also increases the risk of traffic accidents. Therefore, road condition surveys and analyses are crucial. Currently, existing defect deduplication methods rely on the static visual features of defects (size, shape, texture), ignoring spatiotemporal correlations (such as vehicle speed, direction, and time intervals between multiple detections). Traditional Euclidean distance / cosine similarity methods struggle to capture defect deformation features (such as crack extension and pothole enlargement). Fixed spatial ranges (such as 5 meters) and similarity thresholds cannot adapt to multiple types of defects (cracks require high positional accuracy, while potholes require dimensional tolerance). Furthermore, they do not consider the differences in viewing angles and confidence fluctuations between different acquisition devices, easily leading to repeated detection of the same defect (such as cracks and potholes). Therefore, a highly accurate road inspection defect deduplication method is urgently needed to address the problems of existing methods relying on static visual features, failing to effectively capture the essential similarity of defects under geometric deformation, and easily detecting the same defect repeatedly. Summary of the Invention

[0003] This application provides a method for deduplicating road inspection defects, which solves the problem that existing defect deduplication methods rely on static visual features of defects, cannot effectively capture the essential similarity of defects under geometric deformation, and are prone to repeatedly detecting the same defect. By using a preset spatiotemporal fusion model to perform spatiotemporal fusion processing on road defect data from different mobile inspection devices, road defect fusion features are obtained. A preset entropy regularization algorithm is used to calculate the optimal transmission distance between the road defect fusion features and generate a cost matrix. Based on the cost matrix, target road defect data is determined and deduplicated. This solves the problem that existing defect deduplication methods rely on static visual features of defects, cannot effectively capture the essential similarity of defects under geometric deformation, and are prone to repeatedly detecting the same defect.

[0004] In a first aspect, embodiments of this application provide a method for deduplicating road inspection defects, the method comprising the following steps:

[0005] Collect road defect data from different mobile inspection devices;

[0006] The road defect data is processed by a pre-defined spatiotemporal fusion model to obtain road defect fusion features.

[0007] The optimal transmission distance between the fused features of the road defects is calculated using a preset entropy regularization algorithm, and a cost matrix is ​​generated.

[0008] Based on the cost matrix, the target road defect data is determined, and the target road defect data is deduplicated.

[0009] Optionally, the step of performing spatiotemporal fusion processing on the road defect data using a preset spatiotemporal fusion model to obtain road defect fusion features includes:

[0010] The road disease data is grouped and processed according to disease type by a preset spatiotemporal fusion model to obtain road disease data corresponding to different disease types.

[0011] For each type of road defect, feature extraction processing is performed on the road defect data corresponding to the different defect types to obtain road defect features;

[0012] The road defect features are subjected to spatiotemporal fusion processing to obtain road defect fusion features.

[0013] Optionally, the spatiotemporal fusion processing of the road defect features to obtain fused road defect features includes:

[0014] The road defect features are subjected to spatiotemporal synchronization processing to obtain spatiotemporally synchronized road defect features;

[0015] The spatiotemporally synchronized road defect features are weighted and fused to obtain road defect fusion features.

[0016] Optionally, the step of calculating the optimal transmission distance of the road defect fusion features using a preset entropy regularization algorithm and generating a cost matrix includes:

[0017] The road defect fusion features are converted into a probability distribution using the kernel density estimation method;

[0018] Based on the probability distribution, the optimal transmission distance between road defect fusion features is calculated using a preset entropy regularization algorithm;

[0019] Based on the optimal transmission distance, a cost matrix is ​​generated.

[0020] Optionally, determining the target road defect data based on the cost matrix includes:

[0021] The optimal transmission distance between the road defect data is determined in the cost matrix;

[0022] The road defect data whose optimal transmission distance between the road defect data is less than the dynamic threshold is selected as the target road defect data.

[0023] Optionally, before selecting the road defect data whose optimal transmission distance between the road defect data is less than a dynamic threshold as the target road defect data, the method further includes:

[0024] Obtain the current disease type corresponding to the road disease data, and determine the corresponding optimal transmission distance threshold based on the current disease type;

[0025] Obtain the environmental parameters of the current inspection environment, and calculate the current environmental compensation factor based on the environmental parameters;

[0026] A dynamic threshold is generated based on the optimal transmission distance threshold and the current environmental compensation factor.

[0027] Optionally, after performing deduplication on the target road defect data, the method further includes:

[0028] When a manually labeled misjudged sample is detected, the model parameters of the spatiotemporal fusion model are adjusted according to the misjudged sample;

[0029] The dynamic threshold is dynamically updated based on the distribution of historical data.

[0030] Secondly, embodiments of this application provide a deduplication device for road inspection defects, the deduplication device for road inspection defects comprising:

[0031] The data acquisition module is used to collect road defect data from different mobile inspection devices;

[0032] The spatiotemporal fusion processing module is used to perform spatiotemporal fusion processing on the road defect data through a preset spatiotemporal fusion model to obtain road defect fusion features;

[0033] The generation module is used to calculate the optimal transmission distance between the fused features of the road defects using a preset entropy regularization algorithm, and to generate a cost matrix;

[0034] The deduplication module is used to determine the target road defect data based on the cost matrix and to perform deduplication processing on the target road defect data.

[0035] Thirdly, embodiments of the present invention provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the method for deduplication of road inspection defects provided in embodiments of the present invention.

[0036] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the method for deduplication of road inspection defects provided in the embodiments of the present invention.

[0037] The above-mentioned solution of this application has the following beneficial effects: It collects road defect data from different mobile inspection devices; it performs spatiotemporal fusion processing on the road defect data using a preset spatiotemporal fusion model to obtain road defect fusion features; it calculates the optimal transmission distance between the road defect fusion features using a preset entropy regularization algorithm and generates a cost matrix; based on the cost matrix, it determines the target road defect data and performs deduplication processing on the target road defect data. By performing spatiotemporal fusion processing on road defect data from different mobile inspection devices using a preset spatiotemporal fusion model to obtain road defect fusion features, calculating the optimal transmission distance between the road defect fusion features using a preset entropy regularization algorithm, generating a cost matrix, and determining the target road defect data based on the cost matrix, and performing deduplication processing on the target road defect data, this solution solves the problem that existing defect deduplication methods rely on the static visual features of defects, cannot effectively capture the essential similarity of defects under geometric deformation, and are prone to repeatedly detecting the same defect.

[0038] Other beneficial effects of this application will be described in detail in the following detailed description section. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A flowchart illustrating a method for deduplicating road inspection defects according to an embodiment of this application;

[0041] Figure 2 A spatiotemporal fusion flowchart of road defect data is provided as an embodiment of this application;

[0042] Figure 3 A flowchart illustrating an optimal transmission distance calculation method provided in an embodiment of this application;

[0043] Figure 4 This is a flowchart of a dynamic threshold determination method provided in an embodiment of this application;

[0044] Figure 5 A flowchart illustrating another method for deduplicating road inspection defects according to an embodiment of this application;

[0045] Figure 6 This is a schematic diagram of the structure of a deduplication device for road inspection defects provided in an embodiment of this application;

[0046] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0047] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0048] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0049] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0050] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0051] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0052] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0053] like Figure 1 As shown, Figure 1 This is a flowchart of a method for deduplicating road inspection defects according to an embodiment of the present invention. The method for deduplicating road inspection defects includes the following steps:

[0054] 101. Collect road defect data from different mobile inspection equipment.

[0055] In this embodiment of the invention, the deduplication method for road inspection defects described above can be applied to a road defect identification platform. This platform can be built on a server-based or distributed architecture and includes a data interface (for sensor or user uploads), a knowledge database, and a knowledge database construction program. The data interface can be used to obtain a first feature vector to be searched, and the knowledge database construction program can be used to construct the knowledge database. This knowledge database is specifically designed to provide additional relational information for the identified data entities, thereby improving the depth of the data identification system's understanding of the content.

[0056] The aforementioned mobile inspection equipment can be used for road defect detection, such as drones and inspection vehicles. These mobile inspection devices are equipped with visible light cameras, LiDAR, GPS, and other advanced technologies.

[0057] The aforementioned road damage data can be information about road damage collected through mobile inspection equipment, including defects such as cracks and potholes.

[0058] It should be noted that synchronous data collection can be used to collect road defects data from different angles and dimensions using different mobile inspection devices, and integrating the road defect data collected by different inspection devices can help with data analysis and processing.

[0059] 102. The road defect data is processed by spatiotemporal fusion using a pre-defined spatiotemporal fusion model to obtain the road defect fusion features.

[0060] In this embodiment of the invention, the aforementioned pre-defined spatiotemporal fusion model can be a spatiotemporal fusion model based on deep learning or constructed using deep learning, such as LSTM, Transformer, etc. The aforementioned LSTM (Long Short-Term Memory) is a time-recurrent neural network specifically designed to solve the long-term dependency problem of recurrent neural networks (RNNs). It dynamically controls the storage and updating of information by introducing memory units and gating mechanisms (forget gate, input gate, output gate). The aforementioned Transformer is a sequence model based on an attention mechanism, mainly used for natural language processing tasks (such as machine translation). The core feature of Transformer is its efficiency improvement through parallel computation. The aforementioned spatiotemporal fusion model can identify the fusion features of road damage data.

[0061] The above-mentioned spatiotemporal fusion processing can be understood as a process of synchronously integrating the time and spatial information of road defect data through a preset spatiotemporal fusion model.

[0062] The aforementioned road defect fusion features can be obtained by performing spatiotemporal fusion of road defect data through a preset spatiotemporal fusion model.

[0063] In one possible implementation, for example, there is a road defect dataset containing multiple time points and multiple spatial locations. Each data point contains road defect data for that time point and spatial location. A preset spatiotemporal fusion model can be used to integrate the road defect data to obtain road defect fusion features.

[0064] Understandably, by performing spatiotemporal fusion processing on road defect data through a pre-defined spatiotemporal fusion model to obtain road defect fusion features, the accuracy of road defect detection can be improved.

[0065] 103. Calculate the optimal transmission distance between road defect fusion features using a preset entropy regularization algorithm, and generate a cost matrix.

[0066] In this embodiment of the invention, the aforementioned preset entropy regularization algorithm can be a pre-set entropy regularization algorithm. The aforementioned entropy regularization algorithm can transform the original problem into a form that can be solved quickly by introducing entropy regularization. The core idea of ​​the entropy regularization algorithm is to improve its computational speed to an engineering-usable level while maintaining the optimal transmission robustness to deformation.

[0067] The Optimal Transport Distance (OT distance) is a metric used to measure the difference between two probability distributions. The core idea of ​​the OT distance is to transform one distribution into another by minimizing the transport distance.

[0068] Furthermore, the fused features of road defects can be viewed as a probability distribution, and the minimum transmission distance between the fused features can be solved using an entropy regularization algorithm. Minimizing the transmission distance can be the minimum total workload or total cost required to transform one distribution into another; the value of the minimum total cost is the optimal transmission distance between the two.

[0069] Specifically, the formula for the optimal transmission distance is:

[0070]

[0071] in, The transmission plan is represented by a matrix; This represents the probability of a feature point from the source distribution being transmitted to the j-th feature point in the target distribution. This represents the cost function, which can be the cost of the source feature points. The unit mass moves to the target feature point The required cost; Represents the regularization coefficient; This indicates entropy regularization.

[0072] The aforementioned cost matrix can be a two-dimensional square matrix, with rows and columns corresponding to all road defects. Each element C[i][j] in the matrix represents the optimal transmission distance (OT distance) between defect i and defect j. The optimal transmission distance measures the minimum transmission cost required to transform the fused features of defect i into the fused features of defect j. The smaller the cost, the more similar the distributions of the two defects in the feature space, and the more likely they are observations of the same defect at different times or from different perspectives.

[0073] 104. Based on the cost matrix, determine the target road defect data and perform deduplication on the target road defect data.

[0074] In this embodiment of the invention, the rows and columns of the aforementioned cost matrix correspond to all road defects, and each element C[i][j] in the matrix represents the optimal transmission distance (OT distance) between defect i and defect j. The optimal transmission distance measures the minimum transmission cost required to transform the fused features of defect i into the fused features of defect j. The smaller the cost, the more similar the distributions of the two defects in the feature space, and the more likely they are observations of the same defect at different times or from different perspectives; the larger the cost, the less similar the distributions of the two defects in the feature space, and the less likely they are observations of the same defect at different times or from different perspectives.

[0075] Furthermore, the road defect data with the lowest cost can be selected from the cost matrix as the target road defect data, and the target road defect data can be deduplicated.

[0076] The aforementioned target road defect data can be duplicated road defect data.

[0077] The deduplication process described above can be a process of deleting or merging duplicate road defect data.

[0078] Furthermore, the road defect data with the highest cost can be selected from the cost matrix as the target road defect data, and new target road defect data can be added. The aforementioned target road defect data can be non-repeating road defect data.

[0079] It should be noted that when the target road defect data is not duplicated, new target road defect data can be added to the defect dataset to improve road maintenance.

[0080] In this embodiment of the invention, road defect data from different mobile inspection devices are collected; the road defect data are spatiotemporally fused using a preset spatiotemporal fusion model to obtain road defect fusion features; the optimal transmission distance between the road defect fusion features is calculated using a preset entropy regularization algorithm, and a cost matrix is ​​generated; based on the cost matrix, target road defect data is determined, and the target road defect data is deduplicated. By using a preset spatiotemporal fusion model to perform spatiotemporal fusion processing on road defect data from different mobile inspection devices to obtain road defect fusion features, calculating the optimal transmission distance between the road defect fusion features using a preset entropy regularization algorithm, generating a cost matrix, and determining the target road defect data based on the cost matrix, the target road defect data is deduplicated. This solves the problem that existing defect deduplication methods rely on the static visual features of defects, cannot effectively capture the essential similarity of defects under geometric deformation, and are prone to repeatedly detecting the same defect.

[0081] It is understood that in the specific implementation of this application, data related to road defects, distance, spatiotemporal data, matrix data, etc. are involved. When the embodiments in this application are applied to specific products or technologies, user permission or consent is required. Furthermore, the collection, use and processing of related data, as well as the training, deployment and invocation of algorithm models, must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0082] Optionally, in the step of performing spatiotemporal fusion processing on road defect data using a preset spatiotemporal fusion model to obtain road defect fusion features, the road defect data can be grouped according to defect type using the preset spatiotemporal fusion model to obtain road defect data corresponding to different defect types; for each defect type, feature extraction processing is performed on the road defect data corresponding to different defect types to obtain road defect features; and the road defect features are subjected to spatiotemporal fusion processing to obtain road defect fusion features.

[0083] In this embodiment of the invention, the aforementioned preset spatiotemporal fusion model can be a spatiotemporal fusion model built based on deep learning or machine learning, such as LSTM, Transformer, etc. The aforementioned preset spatiotemporal fusion model can extract spatiotemporally invariant features of road defect data and construct a dynamic graph structure. The dynamic graph structure includes node features and edge features. The aforementioned node features fuse visual features, GPS coordinates, timestamps, and vehicle speed. The aforementioned edge features are generated based on a spatiotemporal proximity matrix; a connection is established between two nodes if the spatiotemporal difference is less than 5 seconds and the spatial Euclidean distance is less than 5 meters, with an edge weight of = Where e represents the natural constant, distance represents the Euclidean distance between two diseased nodes, and σ represents the scale parameter. .

[0084] The above-mentioned types of defects can be various types of damage that occur during the use of roads, such as road surface cracks, potholes, and subsidence.

[0085] The above grouping process can be a process of grouping road defect data according to defect type.

[0086] The aforementioned feature extraction process involves analyzing road defect data to extract features that reflect the characteristics of road defects. DeformableResNet-50 can be used as the backbone network, employing deformable convolution kernels to extract variable convolutional features from the road defect data. Deformable convolution kernels can adaptively capture local deformation features of irregular shapes such as cracks and network cracks. The deformable convolution kernel offset Δp is generated through learnable parameters.

[0087]

[0088] in, Indicates the output feature value; The weight of the k-th sampling point is represented by x; x represents the input feature map. Indicates the current position; pk is a fixed sampling point, Δpk is a dynamic offset, which enhances the adaptability to crack deformation and outputs a 256-dimensional visual feature vector (including attributes such as position, size, and texture).

[0089] The core of variable convolution lies in enabling the sampling points of the convolution kernel to adaptively adjust their positions based on the input features, thereby better conforming to the actual shape of the target. For this purpose, a standard convolutional layer, called an offset generation layer, is needed to learn dynamic offsets. The structure of the offset generation layer involves inserting a parallel offset generation branch at a specific stage of the backbone network (such as Deformable ResNet-50) (e.g., before the 3x3 convolutional layer of each residual block). This branch is typically also a convolutional layer. The parameters of the offset generation layer are that its kernel size is usually the same as the main convolutional layer (e.g., 3x3). The input to the offset generation layer is the feature map from the previous layer, and the output has 2N channels, where N is the number of sampling points in the main convolutional layer (N=9 for a 3x3 convolution). These 2N channels correspond to the two-dimensional offset (Δx, Δy) of each of the N sampling points. The offset generation layer is initialized by setting the initial values ​​of the offsets to zero, meaning that the initial sampling point positions are consistent with the standard convolution. The introduction of modulation scalars (ModulationScale) in the offset generation layer (a core idea of ​​DCNv2) is a feature of the more advanced Deformable Convolution v2 (DCNv2). In addition to the offset Δpk, a modulation scalar Δmk (typically ranging from 0 to 1) is learned for each sampling point. The modulation scalar Δmk can be understood as a weight representing the importance of that sampling point. For example, if a sampling point, after offset, falls into a background area unrelated to the disease, the network can learn to set Δmk close to 0, thereby reducing its impact. The number of output channels in the offset generation layer becomes 3N (2N offsets + N modulation scalars).

[0090] The formula for deformable convolution has evolved to:

[0091]

[0092] in, Indicates the output feature value; The weight of the k-th sampling point is represented by x; x represents the input feature map. Δpk represents the current position; pk is a fixed sampling point; Δpk is the dynamic offset; Δmk represents the modulation scalar.

[0093] The aforementioned road damage characteristics can reflect indicators of road damage data, such as the length of cracks and the depth of potholes.

[0094] The aforementioned spatiotemporal fusion processing can be a process of fusing time-series information from road defect features with spatial information from images. Specifically, it can fuse road defect features, GPS coordinates, timestamps, and vehicle speed (format: [longitude, latitude, Δt, direction angle]).

[0095] The aforementioned integrated characteristics of road defects can more comprehensively describe and identify road defects.

[0096] It's understandable that cracks are typically long, thin, and curved, while pits are mostly irregular blocks or holes. For cracks, crack continuity constraints can be added to the loss function of the spatiotemporal fusion model. For example, during training, the offset directions of adjacent sampling points can be encouraged to be smoother, avoiding excessive distortion and allowing them to better fit the crack's direction. For crack detection tasks, directional enhancement can be performed during data preprocessing, such as simulating cracks at different angles, so that the Δpk learned by the spatiotemporal fusion model can better adapt to various directional changes in the crack. For pits, the pit's boundary is more important than its internal texture. Network structure design can guide the offset generation layer to focus more on edge features. For example, low-level features (containing more details and edge information) can be introduced into the offset calculation. Since pits vary in size, multi-scale design in the offset generation branch is particularly important to ensure the model can perceive pit contours at different scales.

[0097] It should be noted that, in order to enable the generated offsets to more accurately capture defects of different sizes (such as fine cracks and larger pits), multi-scale feature fusion can be introduced into the offset generation branch. In the offset generation branch, convolutional layers with different dilation rates are used to process features in parallel, and then the results are fused. Finally, the offset and modulation scalar are generated through a convolutional layer, which helps the network consider the contextual information under different receptive fields when calculating the offset.

[0098] like Figure 2 As shown, Figure 2 This is a flowchart of spatiotemporal fusion of road defect data provided in an embodiment of the present invention, including the following steps:

[0099] 201. Original image.

[0100] The original image can be road damage data.

[0101] 202. Variable convolution feature extraction.

[0102] Among them, DeformableResNet-50 can be used as the backbone network, and Deformable Convolution kernels can be used to extract variable convolution features from the original image.

[0103] 203. Spatiotemporal graph construction.

[0104] The spatiotemporal graph includes spatiotemporal node features and edge features. Spatiotemporal node features are fused with visual features Fv, GPS coordinates, timestamps, and vehicle speed. Edge features are generated based on the spatiotemporal adjacency matrix; a connection is established between two nodes if the time difference is less than 5 seconds and the spatial Euclidean distance is less than 5 meters, with an edge weight of = Where e represents the natural constant, distance represents the Euclidean distance between two diseased nodes, and σ represents the scale parameter. .

[0105] 204. Graph Convolution Aggregation.

[0106] Among them, graph convolution aggregation can be used to output road defect fusion features.

[0107] Optionally, in the step of performing spatiotemporal fusion processing on road defect features to obtain road defect fusion features, the road defect features can be spatiotemporally synchronized to obtain spatiotemporally synchronized road defect features; and the spatiotemporally synchronized road defect features can be weighted and fused to obtain road defect fusion features.

[0108] In this embodiment of the invention, the aforementioned spatiotemporal synchronization processing can be a process of processing road defect features at different times and spatial locations to ensure consistency of road defect features in both time and spatial coordinates. Specifically, spatiotemporally synchronized road defect features can be obtained by combining the GPS coordinates, vehicle speed, and detection timestamp of the road defect features. Spatiotemporally synchronized road defect features can be represented as follows:

[0109] V spatio−temporal =[Longitude, Latitude, Vehicle Speed, Δt, Direction of Travel]

[0110] Among them, longitude and latitude are the geographical coordinates of the disease detection point; vehicle speed represents the speed of the inspection vehicle at the time of detection, which is used to compensate for the positional deviation caused by vehicle movement; Δt represents the time difference between two detections, which is used to measure the time interval of disease detection; and driving direction represents the direction angle of vehicle driving, which is used to determine the direction of vehicle driving trajectory and help predict possible changes in the location of the disease.

[0111] Furthermore, the spatiotemporal drift distance between defects can be calculated using the Haversine formula, dynamically compensating for vehicle motion and improving the accuracy of spatial matching. The purpose of calculating the spatiotemporal drift distance between defects is to calculate the dynamic positional deviation between two defects. The input parameters for the spatiotemporal drift distance are longitude, latitude, vehicle speed, Δt, and direction of travel. The spatiotemporal drift distance can be used to determine whether two defects are the same defect observed at different times. The Haversine formula is a mathematical method for calculating the great circle distance between two points on the Earth's surface, using latitude and longitude to calculate the arc length between the two points.

[0112] Specifically, spatiotemporal drift compensation can be expressed as:

[0113] def spatio_temporal_compensation(gps1, gps2, t1, t2, velocity):

[0114] # Havesing formula for calculating geographical distance

[0115] distance = haversine(gps1, gps2)

[0116] time_diff = abs(t1 - t2)

[0117] # Dynamic compensation for vehicle displacement

[0118] projected_drift = distance - velocity * time_diff

[0119] return projected_drift

[0120] The aforementioned weighted fusion processing can be a process of assigning weights to the spatiotemporally synchronized road defect features according to the confidence levels of the corresponding inspection equipment, and then performing weighted fusion on the road defect features. It is understandable that the data quality collected by different inspection equipment varies. For example, the overhead view from a drone is affected by lighting and cloud cover, resulting in lower image resolution; while the side view from a vehicle-mounted camera is affected by vehicle bumps and obstructions, but has higher resolution.

[0121] Furthermore, a confidence weight can be assigned to the road defect features collected by each device, such as a weight of ≤0.3 for drones and ≥0.7 for vehicle-mounted cameras. The road defect features from multiple devices are then summed according to their confidence weights to obtain the fused road defect features.

[0122] In this embodiment of the invention, the inspection equipment can be equipped with a spatiotemporally coded metasurface array, which is deployed on the roof of the inspection vehicle. The spatiotemporally coded metasurface array integrates a visible light camera, LiDAR, and GPS / IMU modules. The aforementioned metasurface units can independently control their reflection coefficients via an FPGA to achieve spatiotemporal alignment of multi-sensor data (such as mapping LiDAR point clouds to image pixel coordinates). The inspection equipment can also employ a Jetson AGX Orin chip, with a built-in 128-core GPU + 12-core CPU, for edge node computation, supporting real-time preprocessing (resolution: 3840×2160@30fps, point cloud density ≥160pt / m²).

[0123] It should be noted that by fusing GPS coordinates, vehicle speed, timestamps, and driving direction, spatiotemporally synchronized road defect features are obtained. The Haversine formula is then used to calculate the spatiotemporal drift distance after compensating for vehicle displacement, replacing the traditional fixed threshold and achieving accurate spatial matching in mobile inspection scenarios. Weights are assigned to road defect features based on the confidence levels of different inspection devices, and these features are then weighted and fused to obtain fused road defect features. This effectively suppresses interference from differences in the viewing angles of inspection devices and improves the system's compatibility with device variations.

[0124] Optionally, in the step of calculating the optimal transmission distance between road defect fusion features and generating a cost matrix using a preset entropy regularization algorithm, the road defect fusion features can be converted into a probability distribution using a kernel density estimation method; based on the probability distribution, the optimal transmission distance between road defect fusion features is calculated using a preset entropy regularization algorithm; and based on the optimal transmission distance, a cost matrix is ​​generated.

[0125] In this embodiment of the invention, the kernel density estimation method described above is a nonparametric probability density estimation method. This method does not assume any particular distribution for the data; instead, it estimates the distribution based on the data itself. The probability distribution can be expressed as:

[0126]

[0127] in, This represents the probability distribution value; n represents the dimension of the feature vector. Represents any point in the feature space; Let represent the i-th element value of the eigenvector; K is the Gaussian kernel function. The Gaussian kernel function is expressed as:

[0128]

[0129] Where, u= - The bandwidth h=0.5.

[0130] The above probability distribution can be understood as a continuous distribution obtained after kernel density estimation of the road defect fusion features. The probability distribution represents the possible distribution of feature values ​​in the feature space.

[0131] The aforementioned pre-defined entropy regularization algorithm can be a pre-set entropy regularization algorithm. This algorithm transforms the original problem into a form that can be solved quickly by introducing entropy regularization. The core idea of ​​the entropy regularization algorithm is to improve its computational speed to an engineering-usable level while maintaining the optimal transport robustness to deformation.

[0132] The optimal transmission distance mentioned above is a metric used to measure the difference between two probability distributions. The core idea of ​​the optimal transmission distance is to transform one distribution into another by minimizing the transmission distance.

[0133] Furthermore, the road defect fusion feature Fv can be mapped to probability distributions μsource and μtarget, and the formula for the optimal transmission distance is expressed as:

[0134]

[0135] in, Indicates the transmission plan; This represents the probability of a feature point from the i-th feature point of the source distribution being transmitted to the j-th feature point of the target distribution. Represents the transmission cost item; The i-th element represents the eigenvector of the μ source and μ target; The j-th element represents the eigenvector of the μ source and μ target; This represents the entropy regularization term; represents the regularization coefficient; KL(·||·) represents the KL divergence, which is used to measure the difference between the transmission plan π and the probability distribution μ_source ⊗ μ_target.

[0136] The aforementioned cost matrix can be a two-dimensional square matrix, with rows and columns corresponding to all road defects. Each element C[i][j] in the matrix represents the optimal transmission distance (OT distance) between defect i and defect j. The optimal transmission distance measures the minimum transmission cost required to transform the fused features of defect i into the fused features of defect j. The smaller the cost, the more similar the distributions of the two defects in the feature space, and the more likely they are observations of the same defect at different times or from different perspectives. Conversely, the larger the cost, the less similar the distributions of the two defects in the feature space, and the less likely they are observations of the same defect at different times or from different perspectives.

[0137] Specifically, the iterative formula is as follows:

[0138] def sinkhorn(u, v, C, lambda_reg, max_iter=100):

[0139] K = np.exp(-C / lambda_reg)

[0140] for _ in range(max_iter):

[0141] u = 1 / (K @ v)

[0142] v = 1 / (KT @ u)

[0143] return np.diag(u) @ K @ np.diag(v)

[0144] like Figure 3 As shown, Figure 3 This is a flowchart of an optimal transmission distance calculation method provided by an embodiment of the present invention, specifically including the following steps:

[0145] 301. Fusion characteristics.

[0146] Among them, the fusion feature is the fusion feature of road defects.

[0147] 302. Probability distribution.

[0148] In this process, the road defects fusion features are mapped to a probability distribution.

[0149] 303. Entropy regularization iteration.

[0150] Among them, based on the probability distribution, the optimal transmission distance between road defect fusion features is accelerated by using an entropy regularization algorithm.

[0151] 304. Optimal transmission distance output.

[0152] The optimal transmission distance between road defect fusion features is calculated using an entropy regularization algorithm.

[0153] It should be noted that the road defect fusion features are converted into a probability distribution by kernel density estimation. Based on the probability distribution, the optimal transmission distance between the road defect fusion features is calculated using a preset entropy regularization algorithm. A cost matrix is ​​generated based on the optimal transmission distance. The cost matrix can accurately measure the similarity between different road defects.

[0154] Optionally, in the step of determining the target road defect data based on the cost matrix, the optimal transmission distance between the road defect data can be determined in the cost matrix; and the road defect data whose optimal transmission distance between the road defect data is less than the dynamic threshold can be selected as the target road defect data.

[0155] In this embodiment of the invention, the rows and columns of the aforementioned cost matrix correspond to all road defects, and each element C[i][j] in the matrix represents the optimal transmission distance (OT distance) between defect i and defect j. The optimal transmission distance between road defect data can be determined from the cost matrix, and road defect data with an optimal transmission distance less than a dynamic threshold can be selected as target road defect data.

[0156] The optimal transmission distance mentioned above can be the minimum transmission metric between road defect data.

[0157] The aforementioned dynamic thresholds can be obtained based on the baseline optimal transmission distance threshold and the environmental compensation factor for the corresponding disease type. The baseline optimal transmission distance threshold can be a pre-set baseline optimal transmission distance threshold for different disease types. The environmental compensation factor can be a dynamically adjusted numerical parameter, used to adjust the compensation value of the environmental threshold.

[0158] The aforementioned target road defect data can be duplicated road defect data.

[0159] In one possible implementation, for example, if the optimal transmission distance between road defect data A and road defect data B is less than a dynamic threshold, then road defect data A and road defect data B can be identified as duplicate road defect data.

[0160] Optionally, before selecting road defect data whose optimal transmission distance is less than the dynamic threshold as the target road defect data, it is also possible to obtain the current defect type corresponding to the road defect data, and determine the corresponding optimal transmission distance threshold based on the current defect type; obtain the environmental parameters of the current inspection environment, and calculate the current environmental compensation factor based on the environmental parameters; and generate a dynamic threshold based on the optimal transmission distance threshold and the current environmental compensation factor.

[0161] In this embodiment of the invention, the aforementioned current disease type can be the disease type corresponding to road disease data, and the disease type can be cracks, potholes, etc.

[0162] The aforementioned optimal transmission distance threshold can be the basic optimal transmission distance threshold corresponding to the disease type.

[0163] The environmental parameters mentioned above can be sunlight, water accumulation, rain, fog, etc.

[0164] The aforementioned current environmental compensation factor can be dynamically calculated based on the environmental parameters of the current inspection environment to adjust the environmental threshold. Specifically, the water depth can be estimated with high precision from the inspection video using a tire geometry inversion algorithm. Specifically, the tire outline can be identified in the inspection video, and the water depth can be inferred from the elliptic distortion rate (accuracy ±1cm). The aforementioned tire geometry inversion algorithm is a calculation method that maps the geometric features of the tire contact surface to the internal structure or motion parameters of the tire using the principle of inversion transformation. The core of the tire geometry inversion algorithm is to utilize the inverse relationship of the distance to an inverted circle (or sphere) to transform the complex deformation of the tire's contact with the ground into a calculable geometric model. The aforementioned elliptic distortion rate can be a parameter describing the degree to which an ellipse or ellipsoid deviates from a circle or sphere.

[0165] Furthermore, the optimal transmission distance threshold can be added to the current environmental compensation factor to obtain the dynamic threshold.

[0166] Specifically, the dynamic threshold can be expressed in Table 1 below:

[0167] Table 1

[0168] Disease types Basic OT threshold Environmental compensation factors crack 0.05 Light intensity × 0.01 potholes 0.12 Water depth × 0.05

[0169] like Figure 4 As shown, Figure 4 This is a flowchart of a dynamic threshold determination method provided by an embodiment of the present invention, including the following steps:

[0170] 401. Optimal transmission distance.

[0171] The optimal transmission distance mentioned above is the minimum transmission distance between road defect data.

[0172] 402. Dynamic threshold determination.

[0173] If the optimal transmission distance is less than the dynamic threshold, it is determined to be duplicate road defect data, and the process proceeds to step 403; otherwise, if the optimal transmission distance is greater than or equal to the dynamic threshold, it is determined to be non-duplicate road defect data, and the process proceeds to step 405.

[0174] 403. Merge records.

[0175] Among them, the record with the highest confidence in the road defect data (the confidence quantification formula is: visual clarity × equipment weight) and the spatiotemporal fusion features of the inspection records of the road defect data are associated with the same defect ID.

[0176] 404. Update history database.

[0177] The merged road defect data will be stored in a historical database.

[0178] 405. New record added.

[0179] Among them, road defect data will be stored in the historical database as a newly added road defect record.

[0180] In one possible implementation, if the optimal transmission distance between road defect data A and road defect data B is less than the basic optimal transmission threshold plus an environmental compensation factor, then road defect data A and road defect data B are determined to be duplicate road defect data. The record with the highest confidence level for road defect data A and road defect data B (confidence quantification formula: visual clarity × device weight) can be selected, and the spatiotemporal fusion features of the inspection records of road defect data A and road defect data B are associated with the same defect ID to construct a spatiotemporal trajectory of the defect, which can be used to analyze defects such as crack extension trends. If the optimal transmission distance between road defect data A and road defect data B is greater than or equal to the basic optimal transmission threshold plus an environmental compensation factor, then road defect data A and road defect data B are determined to be non-duplicate road defect data. Independent defect records can be added for road defect data A and road defect data B respectively, and new defect IDs can be assigned.

[0181] Optionally, after the deduplication process of the target road defect data, when a manually marked misjudged sample is detected, the model parameters of the spatiotemporal fusion model are adjusted according to the misjudged sample; and the dynamic threshold is dynamically updated according to the historical data distribution.

[0182] In this embodiment of the invention, the aforementioned misclassified samples can be understood as samples that are incorrectly classified during data analysis. Misclassified samples include false positives and false negatives. A false positive can be a sample pair that the system classifies as "duplicate" but is manually confirmed as "different diseases." A false negative can be a sample pair that the system classifies as "different" but is manually confirmed as "the same disease."

[0183] The above parameter adjustments can be a process of improving model performance by addressing the model parameters of the spatiotemporal fusion model through misjudged samples.

[0184] Specifically, when a manually labeled misclassified sample is detected, the model parameters are fine-tuned:

[0185]

[0186] in, This represents the optimal transmission distance, which can be understood as the model's predicted output (part of the loss) for misjudged samples based on the current road defect data. These are the current parameters of the model. These are the parameters before fine-tuning (or the parameters from the previous iteration); It is the elasticity coefficient, which controls the update amplitude of the parameter; It is the squared L2 norm of the parameter change, which is a regularization term to prevent the parameter from being updated too quickly or too large.

[0187] The aforementioned historical data distribution can be a set of statistical information accumulated by the system during its historical operation regarding the optimal transmission distance for each disease type that is identified as the same disease.

[0188] The aforementioned dynamic update can be a process of modifying dynamic thresholds in real time during system runtime.

[0189] Specifically, the dynamic threshold can be updated based on the distribution of historical data:

[0190]

[0191] in, This represents the average optimal transmission distance for similar diseases; It represents the standard deviation.

[0192] like Figure 5 As shown, Figure 5 This is a flowchart of a method for deduplicating road inspection defects according to an embodiment of the present invention, including:

[0193] 501. Manual sample correction.

[0194] Among these, maintenance personnel discovered and flagged system errors during system use. 502. Error Analysis.

[0195] Among these, the process involves "reviewing" and "diagnosing" manually labeled misjudged samples.

[0196] 503. Model fine-tuning.

[0197] Among these, the model parameters are adjusted based on the misjudged samples.

[0198] 504. Dynamic threshold update.

[0199] Among them, the dynamic threshold is adjusted based on the distribution of historical data.

[0200] In this embodiment of the invention, when maintenance personnel mark misjudged samples, the invention triggers fine-tuning of model parameters to improve the robustness and accuracy of disease feature extraction; for each disease type, the dynamic threshold can be updated according to the historical data distribution, which can improve the accuracy of disease identification.

[0201] like Figure 6 As shown, this embodiment of the invention provides a deduplication device for road inspection defects, which includes:

[0202] The data acquisition module 601 is used to collect road defect data from different mobile inspection devices;

[0203] The spatiotemporal fusion processing module 602 is used to perform spatiotemporal fusion processing on the road defect data through a preset spatiotemporal fusion model to obtain road defect fusion features;

[0204] The generation module 603 is used to calculate the optimal transmission distance between the fused features of the road defects using a preset entropy regularization algorithm, and to generate a cost matrix;

[0205] The deduplication module 604 is used to determine the target road defect data based on the cost matrix and to perform deduplication processing on the target road defect data.

[0206] Optionally, the spatiotemporal fusion processing module 602 is further configured to group the road defect data according to defect type using a preset spatiotemporal fusion model to obtain road defect data corresponding to different defect types; for each defect type, perform feature extraction processing on the road defect data corresponding to the different defect types to obtain road defect features; and perform spatiotemporal fusion processing on the road defect features to obtain road defect fusion features.

[0207] Optionally, the spatiotemporal fusion processing module 602 is further configured to perform spatiotemporal synchronization processing on the road defect features to obtain spatiotemporally synchronized road defect features; and to perform weighted fusion processing on the spatiotemporally synchronized road defect features to obtain road defect fusion features.

[0208] Optionally, the generation module 603 is further configured to convert the road defect fusion features into a probability distribution using a kernel density estimation method; calculate the optimal transmission distance between the road defect fusion features based on the probability distribution using a preset entropy regularization algorithm; and generate a cost matrix based on the optimal transmission distance.

[0209] Optionally, the deduplication module 604 is further configured to determine the optimal transmission distance between the road defect data in the cost matrix; and select the road defect data whose optimal transmission distance between the road defect data is less than a dynamic threshold as the target road defect data.

[0210] Optionally, the device is further configured to acquire the current disease type corresponding to the road disease data, and determine the corresponding optimal transmission distance threshold based on the current disease type; acquire the environmental parameters of the current inspection environment, and calculate the current environmental compensation factor based on the environmental parameters; and generate a dynamic threshold based on the optimal transmission distance threshold and the current environmental compensation factor.

[0211] Optionally, the device is also used to adjust the model parameters of the spatiotemporal fusion model according to the misjudged sample when a manually labeled misjudged sample is detected; and to dynamically update the dynamic threshold according to the historical data distribution.

[0212] like Figure 7 As shown, this embodiment of the invention also provides an electronic device, including a processor, which can execute any of the above-described methods for deduplication of road inspection defects.

[0213] Specifically, it includes a processor 701 and a memory 702, as well as a computer program stored in the memory 702 and capable of running on the processor 701, which executes a method for removing duplicates from road inspection defects, wherein:

[0214] Processor 701 executes the calculator program for the deduplication method of road inspection defects stored in memory 702, and performs the following steps:

[0215] Collect road defect data from different mobile inspection devices;

[0216] The road defect data is processed by a pre-defined spatiotemporal fusion model to obtain road defect fusion features.

[0217] The optimal transmission distance between the fused features of the road defects is calculated using a preset entropy regularization algorithm, and a cost matrix is ​​generated.

[0218] Based on the cost matrix, the target road defect data is determined, and the target road defect data is deduplicated.

[0219] Optionally, the process executed by processor 701 to perform spatiotemporal fusion processing on the road defect data using a preset spatiotemporal fusion model to obtain road defect fusion features includes:

[0220] The road disease data is grouped and processed according to disease type by a preset spatiotemporal fusion model to obtain road disease data corresponding to different disease types.

[0221] For each type of road defect, feature extraction processing is performed on the road defect data corresponding to the different defect types to obtain road defect features;

[0222] The road defect features are subjected to spatiotemporal fusion processing to obtain road defect fusion features.

[0223] Optionally, the spatiotemporal fusion processing of the road defect features performed by the processor 701 to obtain fused road defect features includes:

[0224] The road defect features are subjected to spatiotemporal synchronization processing to obtain spatiotemporally synchronized road defect features;

[0225] The spatiotemporally synchronized road defect features are weighted and fused to obtain road defect fusion features.

[0226] Optionally, the processor 701 executes the step of calculating the optimal transmission distance between the fused road defect features using a preset entropy regularization algorithm and generating a cost matrix, including:

[0227] The road defect fusion features are converted into a probability distribution using the kernel density estimation method;

[0228] Based on the probability distribution, the optimal transmission distance between road defect fusion features is calculated using a preset entropy regularization algorithm;

[0229] Based on the optimal transmission distance, a cost matrix is ​​generated.

[0230] Optionally, the process executed by processor 701 to determine the target road defect data based on the cost matrix includes:

[0231] The optimal transmission distance between the road defect data is determined in the cost matrix;

[0232] The road defect data whose optimal transmission distance between the road defect data is less than the dynamic threshold is selected as the target road defect data.

[0233] Optionally, before selecting the road defect data whose optimal transmission distance between the road defect data is less than a dynamic threshold as the target road defect data, the method executed by the processor 701 further includes:

[0234] Obtain the current disease type corresponding to the road disease data, and determine the corresponding optimal transmission distance threshold based on the current disease type;

[0235] Obtain the environmental parameters of the current inspection environment, and calculate the current environmental compensation factor based on the environmental parameters;

[0236] A dynamic threshold is generated based on the optimal transmission distance threshold and the current environmental compensation factor.

[0237] Optionally, after the deduplication process is performed on the target road defect data, the method executed by the processor 701 further includes:

[0238] When a manually labeled misjudged sample is detected, the model parameters of the spatiotemporal fusion model are adjusted according to the misjudged sample;

[0239] The dynamic threshold is dynamically updated based on the distribution of historical data.

[0240] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the deduplication method for road inspection defects provided in this invention and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0241] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for deduplicating road inspection defects, characterized in that, The method includes the following steps: Collect road defect data from different mobile inspection devices; The road defect data is processed by a pre-defined spatiotemporal fusion model to obtain road defect fusion features. The optimal transmission distance between the fused features of the road defects is calculated using a preset entropy regularization algorithm, and a cost matrix is ​​generated. Based on the cost matrix, the target road defect data is determined, and the target road defect data is deduplicated.

2. The method for removing duplicate road inspection defects as described in claim 1, characterized in that, The process of performing spatiotemporal fusion processing on the road defect data using a preset spatiotemporal fusion model to obtain road defect fusion features includes: The road disease data is grouped and processed according to disease type by a preset spatiotemporal fusion model to obtain road disease data corresponding to different disease types. For each type of road defect, feature extraction processing is performed on the road defect data corresponding to the different defect types to obtain road defect features; The road defect features are subjected to spatiotemporal fusion processing to obtain road defect fusion features.

3. The method for removing duplicate road inspection defects as described in claim 2, characterized in that, The process of performing spatiotemporal fusion processing on the road defect features to obtain fused road defect features includes: The road defect features are subjected to spatiotemporal synchronization processing to obtain spatiotemporally synchronized road defect features; The spatiotemporally synchronized road defect features are weighted and fused to obtain road defect fusion features.

4. The method for removing duplicate road inspection defects as described in claim 1, characterized in that, The step of calculating the optimal transmission distance between the fused features of road defects using a preset entropy regularization algorithm and generating a cost matrix includes: The road defect fusion features are converted into a probability distribution using the kernel density estimation method; Based on the probability distribution, the optimal transmission distance between road defect fusion features is calculated using a preset entropy regularization algorithm; Based on the optimal transmission distance, a cost matrix is ​​generated.

5. The method for removing duplicate road inspection defects as described in claim 1, characterized in that, The determination of target road defect data based on the cost matrix includes: The optimal transmission distance between the road defect data is determined in the cost matrix; The road defect data whose optimal transmission distance between the road defect data is less than the dynamic threshold is selected as the target road defect data.

6. The method for removing duplicate road inspection defects as described in claim 5, characterized in that, Before selecting the road defect data whose optimal transmission distance between them is less than a dynamic threshold as the target road defect data, the method further includes: Obtain the current disease type corresponding to the road disease data, and determine the corresponding optimal transmission distance threshold based on the current disease type; Obtain the environmental parameters of the current inspection environment, and calculate the current environmental compensation factor based on the environmental parameters; A dynamic threshold is generated based on the optimal transmission distance threshold and the current environmental compensation factor.

7. The method for removing duplicate road inspection defects as described in claim 1, characterized in that, After performing deduplication on the target road defect data, the method further includes: When a manually labeled misjudged sample is detected, the model parameters of the spatiotemporal fusion model are adjusted according to the misjudged sample; The dynamic threshold is dynamically updated based on the distribution of historical data.

8. A deduplication device for road inspection defects, characterized in that, The deduplication device for road inspection defects includes: The data acquisition module is used to collect road defect data from different mobile inspection devices; The spatiotemporal fusion processing module is used to perform spatiotemporal fusion processing on the road defect data through a preset spatiotemporal fusion model to obtain road defect fusion features; The generation module is used to calculate the optimal transmission distance between the fused features of the road defects using a preset entropy regularization algorithm, and to generate a cost matrix; The deduplication module is used to determine the target road defect data based on the cost matrix and to perform deduplication processing on the target road defect data.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method for deduplication of road inspection defects as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the deduplication method for road inspection defects as described in any one of claims 1 to 7.

Images