Order distribution maintenance method, system and equipment for intelligent patrol disease maintenance, medium and product

Abstract

The invention discloses a dispatch maintenance method, system, device, medium and product for intelligent patrol disease maintenance, and relates to the field of road anomaly detection and road management and maintenance, the method comprises the following steps: collecting a road image of a current road section, and obtaining geographical location information of the road image; inputting the road image into a deep convolutional neural network cluster to obtain road disease information in the road image, including the category of the disease, the severity level of the disease and the pixel coordinate of the disease; and judging whether maintenance is needed or not according to the severity level of the disease in the road disease information, and if maintenance is needed, inputting the road disease information and the geographic position information of the road image into a trained deep Q network model, generating an order dispatching decision strategy and dispatching an order. According to the invention, an automatic process from road patrol and rapid and accurate acquisition of disease information to order maintenance and timely feedback is realized.

Description

Technical Field

[0001] The present application relates to the field of road anomaly detection and road maintenance, and in particular to a maintenance method, system, equipment, medium and product for dispatching orders for intelligent inspection and repair of defects. Background Art

[0002] Traditional road defect inspections mainly rely on manual inspections, which is not only inefficient, but also has high subjectivity and error rates. At the same time, there are safety hazards for inspectors when working.

[0003] At present, there are some smart inspection devices on the market that replace manual inspections and realize intelligent inspections. However, whether it is manual inspections or smart inspections, the diseases reported by the inspections rely on manual dispatching and need to be reviewed manually. Manual dispatching is still inefficient and has a great impact on the real-time and accuracy of maintenance dispatching. With the rapid development of deep learning and artificial intelligence technology, deep reinforcement learning can learn from historical data, replace manual decision-making, and realize the automated process from inspection to dispatching. Summary of the invention

[0004] The purpose of this application is to provide a maintenance method, system, equipment, medium and product for dispatching orders for intelligent inspection and repair of defects, which can realize an automated process from inspection to dispatching orders, improve efficiency and reduce errors caused by human operation.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a maintenance method for dispatching orders for intelligent inspection and repair of defects, including:

[0007] Collecting a road image of the current road section and obtaining geographic location information of the road image;

[0008] Inputting the road image into a deep convolutional neural network cluster to obtain road disease information in the road image, including the type of disease, the severity level of the disease, and the pixel coordinates of the disease;

[0009] Determine whether maintenance is required based on the severity of the road damage information; if maintenance is required, input the road damage information and the geographical location information of the road image into the trained deep Q network model, generate a dispatch decision strategy, and perform dispatch;

[0010] If maintenance is not required, the road image of the next road section is collected, the next road section is used as the current road section and the above steps are repeated.

[0011] Optionally, the road damage information and the geographical location information of the road image are input into a trained deep Q network model to generate a dispatch decision strategy and perform dispatch, specifically including:

[0012] Inputting state information into the trained deep Q network model to calculate the Q values ​​of all actions; the state information includes road disease information and geographical location information of the road image;

[0013] The Q values ​​of all actions are compared and the action corresponding to the largest Q value is selected as the optimal action to generate a dispatch decision strategy, which includes the maintenance unit to which it belongs, the dispatch time and the priority.

[0014] Optionally, the road damage information and the geographical location information of the road image are input into a trained deep Q network model to generate a dispatch decision strategy and perform dispatch, which also includes:

[0015] Collect historical dispatch data, including status information, dispatch decision strategy and actual maintenance effect. The actual maintenance effect includes: repair progress, repair quality, repair time and feedback information. Pre-process the historical dispatch data to generate state-action-reward-new state four-tuple data; the state is the state information, the action is the dispatch decision strategy, the reward is the reward value determined according to the actual maintenance effect, and the new state is the state information updated based on the feedback information;

[0016] Constructing a deep Q network model, wherein the deep Q network model includes a current deep Q network model and a target deep Q network model; the current deep Q network model is used to generate current Q values ​​of all actions according to the input state information, and select the action corresponding to the maximum Q value as the dispatch decision strategy output;

[0017] Establish an experience replay buffer to store the state-action-reward-new state quadruple data;

[0018] During the training process, randomly extracting state-action-reward-new state quadruple data from the experience replay buffer to train the current deep Q network model;

[0019] Acquire the network structure of the current deep Q network, construct a target deep Q network having the network structure of the current deep Q network, the target deep Q network, acquire the network parameters of the current deep Q network at fixed time intervals, and update the network parameters of the target deep Q network in a delayed update manner;

[0020] The error between the current deep Q network model and the target deep Q network model is calculated using the mean square error loss function, and the parameters of the current deep Q network model are updated by minimizing the mean square error loss function through the gradient descent method to obtain the trained deep Q network model.

[0021] Optionally, the expression of the target Q value generated by the target deep Q network model is:

[0022]

[0023] Where Q2 represents the target Q value, r represents the immediate reward, γ represents the discount factor, s′ represents the next state information of the current state information corresponding to the current Q value, a′ represents an action corresponding to the Q value of the next state information, A represents the set of all actions corresponding to the Q value of the next state information, and θ i - Represents the parameters of the target deep Q-network model.

[0024] Optionally, the road damage information and the geographical location information of the road image are input into a trained deep Q network model, a dispatch decision strategy is generated and dispatch is performed, and then the following steps are further included:

[0025] Obtain the actual maintenance effect after dispatching orders according to the dispatch decision strategy;

[0026] Update the reward according to the actual maintenance effect, and generate a new state-action-reward-new state quadruple;

[0027] The new state-action-reward-new-state quadruple data is stored in the experience replay buffer, and the trained deep Q network model is incrementally trained based on the new state-action-reward-new-state quadruple data.

[0028] In the second aspect, the present application provides a maintenance system for intelligent inspection and repair of defects, including:

[0029] The vehicle-mounted wide-angle camera is installed outside the vehicle and connected to the central industrial computer to collect road images of the current section;

[0030] A global navigation satellite system positioning antenna, mounted on the vehicle-mounted wide-angle camera and connected to a central industrial control computer, for obtaining geographic location information of the road image;

[0031] The central industrial control computer is installed in the vehicle and includes a network module and a server. It is used to generate a dispatch decision strategy and perform dispatch based on the road image of the current road section captured by the vehicle-mounted wide-angle camera and the geographic location information of the road image acquired by the global navigation satellite system positioning antenna, using the above-mentioned dispatch maintenance method for intelligent inspection and disease repair.

[0032] Optionally, the maintenance system for dispatching orders for intelligent inspection and repair of defects further includes:

[0033] A rotary encoder is installed on the wheel and connected to the central industrial computer. It is used to send a pulse signal generated by the rotation of the wheel during its forward movement to the central industrial computer. The central industrial computer calculates the distance the vehicle has traveled based on the pulse signal. Every time the vehicle travels a preset distance, the central industrial computer sends a command to the on-board wide-angle camera to capture an image.

[0034] In a third aspect, the present application provides a computer device comprising: 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 of any of the above-mentioned methods for dispatching maintenance for smart inspection and defect repair.

[0035] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-mentioned methods for dispatching maintenance orders for smart inspection and defect repair.

[0036] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned methods for dispatching maintenance for intelligent inspection and defect repair.

[0037] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0038] The present application provides a maintenance method and system for dispatching orders for intelligent inspection and repair of road defects. By using lightweight road inspection equipment to collect road defect information, the inspection efficiency is improved, the dependence on human resources is reduced, and the false alarms and missed reports that often occur in traditional manual inspections are effectively avoided. Especially when dealing with large-scale and high-frequency inspection tasks, the delayed response and processing time can be greatly reduced, thereby realizing the rapid and accurate acquisition and immediate feedback of defect information, and providing timely and reliable data support for road maintenance decisions; by applying deep reinforcement learning algorithms to the collected defect information for intelligent dispatching decisions, it can automatically analyze factors such as the type, severity and geographical location of the defect, scientifically and reasonably generate maintenance work orders, and accurately push them to the corresponding road maintenance departments. Not only does it overcome the problem of inconsistent and inaccurate evaluation results caused by subjective judgment and quality differences of personnel in the traditional manual inspection and dispatching process, but it also optimizes resource allocation through algorithms, solves the problem of uneven resource allocation and insufficient monitoring in some areas that may be caused by manual operation, thereby realizing the accurate dispatch and efficient execution of maintenance tasks, improving the overall efficiency and effect of road maintenance, and ensuring the safety and smooth operation of road facilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 This is an application environment diagram of a maintenance method for dispatching orders for intelligent inspection and repair of defects in one embodiment of the present application;

[0041] Figure 2 A flowchart of a maintenance method for dispatching orders for intelligent inspection and repair of defects provided in one embodiment of the present application;

[0042] Figure 3 A control relationship diagram of a maintenance system for dispatching orders for intelligent inspection and repair of defects provided in one embodiment of the present application;

[0043] Figure 4 A working diagram of a maintenance system for intelligent inspection and repair of defects provided in one embodiment of the present application;

[0044] Figure 5 A schematic diagram of a detailed implementation process of a maintenance method for dispatching orders for intelligent inspection and repair of defects provided in one embodiment of the present application;

[0045] Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application.

[0046] Figure numerals: 1-central industrial control computer, 2-vehicle-mounted wide-angle camera, 3-global navigation satellite system positioning antenna, 4-rotary encoder. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0048] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0049] The maintenance method for intelligent inspection and repair of defects provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the road image of the current section and the geographical location information of the road image to the server 104. After the server 104 receives the road image of the current section and the geographical location information of the road image, for the road image of the current section and the geographical location information of the road image, the server 104 inputs the road image into the deep convolutional neural network cluster to obtain the road disease information in the road image, including the category of the disease, the severity level of the disease, and the pixel coordinates of the disease; judge whether maintenance is required according to the severity level of the disease in the road disease information, if maintenance is required, input the road disease information and the geographical location information of the road image into the trained deep Q network model, generate a dispatch decision strategy and dispatch; if maintenance is not required, collect the road image of the next section, take the next section as the current section and repeat the above steps. The server 104 may feed back the obtained dispatch decision strategy to the terminal 102. In addition, in some embodiments, the dispatch maintenance method for intelligent inspection and repair of defects may also be implemented by the server 104 or the terminal 102 alone, such as the terminal 102 may directly process the road image of the current road section and the geographical location information of the road image, or the server 104 may obtain the road image of the current road section and the geographical location information of the road image from the data storage system, and process the road image of the current road section and the geographical location information of the road image.

[0050] In an exemplary embodiment, Figure 2 As shown, a flow chart of a maintenance method for dispatching orders for intelligent inspection and repair of defects is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following steps 201 to 203. Among them:

[0051] Step 201 : collect a road image of a current road section and obtain geographical location information of the road image.

[0052] Step 202: Input the road image into a deep convolutional neural network cluster to obtain road damage information in the road image, including the type of damage, the severity level of the damage, and the pixel coordinates of the damage.

[0053] Step 203, judging whether maintenance is needed according to the severity level of the road damage information, and if maintenance is needed, inputting the road damage information and the geographic location information of the road image into the trained deep Q network model, generating a dispatch decision strategy and performing dispatch.

[0054] The above steps 201 to 203 are implemented to efficiently collect road damage data, automatically pre-process and intelligently make dispatch decisions.

[0055] In this implementation, in order to collect the road image of the current road section and obtain the geographical location information of the road image, step 201 specifically includes:

[0056] Before collecting the road image of the current road section, the vehicle-mounted wide-angle camera 2 is first calibrated with external parameters using a chessboard;

[0057] As the vehicle starts, the on-board wide-angle camera 2 starts to collect the road image of the current road section, and the global navigation satellite system positioning antenna 3 obtains the geographical location information of the road image. In combination with the above-mentioned external parameter calibration, the perspective transformation matrix of the road image of the current road section can be obtained. The perspective transformation matrix can be used to map the collected road image of the current road section into a top view, and the GPS coordinates of any pixel position on the road image of the current road section can be calculated based on the correspondence between the pixel coordinates of the top view and the world coordinates.

[0058] In another exemplary embodiment of the present application, in order to input the road damage information and the geographical location information of the road image into the trained deep Q network model, generate a dispatch decision strategy and perform dispatch, specifically including:

[0059] The state information is input into the trained deep Q network model to calculate the Q value of all actions; the state information includes road damage information and geographical location information of the road image. Specifically, the state information is a comprehensive description of the current road damage status, including: the type of damage (such as potholes, cracks, subsidence, etc.), the severity level of the damage (from mild to severe), the pixel coordinates of the damage (i.e. the location of the damage in the image), and the geographical location information (such as longitude and latitude, road name). This information is integrated into a state vector and input into the deep Q network model. Among them, the category of defects is defined according to the "Highway Technical Condition Assessment Standard" (JTG 5210-2018). At the same time, the "Highway Technical Condition Assessment Standard" also defines the severity level of defects, which is mainly based on size. The size depends on camera calibration and segmentation model in the network cluster for calculation. These algorithms are built into lightweight intelligent detection equipment. The detector detects the defects and performs instance segmentation. Then the segmented mask image is perspective transformed. The perspective mask can calculate the size information of the defect based on the correspondence between the pixels at the time of calibration and the actual size. The pixel coordinates of the defect are the results of the detector. The geographic location information refers to the collected vehicle-mounted GPS, and the name of the road can be obtained by spatial distance matching with the local road information data.

[0060] The structure of the deep Q network model includes:

[0061] State input layer: The input state vector is processed by several hidden layers, usually including convolutional layers (for extracting spatial features) and fully connected layers (for further abstracting features).

[0062] Action space: The action space defines all possible actions that can be taken. For example, determining the dispatch time (immediate dispatch or delayed dispatch), setting the priority of the maintenance task.

[0063] According to the Q value, the action corresponding to the maximum Q value is selected as the optimal action to generate a dispatch decision strategy, which includes the maintenance unit, dispatch time and priority. For each possible action, the deep Q network model will output a Q value, which represents the expected return (i.e. long-term benefit) of executing the action in the current state. The Q values ​​of all actions are compared and the action with the largest Q value is selected as the optimal dispatch decision strategy in the current state.

[0064] In another exemplary embodiment of the present application, in order to optimize the deep Q network model, reduce false positives and false negatives, and improve the accuracy of dispatching decisions, the road damage information and the geographic location information of the road image are input into the deep Q network model, a dispatching decision strategy is generated, and dispatching is performed, which also includes:

[0065] Collect historical dispatch data, including status information, dispatch decision strategy and actual maintenance effect. The actual maintenance effect includes: repair progress, repair quality, repair time and feedback information. Pre-process the historical dispatch data to generate state-action-reward-new state four-tuple data; the state is the state information, the action is the dispatch decision strategy, the reward is the reward value determined according to the actual maintenance effect, and the new state is the state information updated based on the feedback information;

[0066] Constructing a deep Q network model, wherein the deep Q network model includes a current deep Q network model and a target deep Q network model; the current deep Q network model is used to generate current Q values ​​of all actions according to the input state information, and select the action corresponding to the maximum Q value as the dispatch decision strategy output;

[0067] Establish an experience replay buffer to store the state-action-reward-new state quadruple data;

[0068] During the training process, randomly extracting state-action-reward-new state quadruple data from the experience replay buffer to train the current deep Q network model;

[0069] Acquire the network structure of the current deep Q network, construct a target deep Q network having the network structure of the current deep Q network, the target deep Q network, acquire the network parameters of the current deep Q network at fixed time intervals, and update the network parameters of the target deep Q network in a delayed update manner;

[0070] The error between the current deep Q network model and the target deep Q network model is calculated using a mean square error loss function, and the parameters of the current deep Q network model are updated by minimizing the mean square error loss function by a gradient descent method to obtain a trained deep Q network model. The target deep Q network model is used to calculate the training error and provide a stable target Q value for the current deep Q network.

[0071] Specifically, the current deep Q network model is used to estimate the Q value of each possible action in the current state and generate actual action decisions. The current deep Q network model updates its weight parameters in each iteration. The structure of the target deep Q network model is the same as the current deep Q network model, but the target deep Q network model weight parameters are updated less frequently or updated through soft updates. The main task of the target deep Q network model is to provide a relatively stable target Q value for the current deep Q network model for calculating the training error.

[0072] During the training of a deep Q network, if the target Q value changes too much with each training step, it may cause the parameters of the current deep Q network model to oscillate violently, leading to unstable training. This is because the target Q value is generated by the network itself, and therefore may be directly affected by the update of the network parameters.

[0073] The expression of the current Q value generated by the current deep Q network model is:

[0074] Q1=Q(s,a,θ i );

[0075] Among them, Q1 represents the current Q value, s represents the current state information, a represents the current action corresponding to the current state information, θ i Represents the parameters of the current deep Q network model.

[0076] The expression of the target Q value generated by the target deep Q network model is:

[0077]

[0078] Where Q2 represents the target Q value, r represents the immediate reward, γ represents the discount factor, s′ represents the next state information of the current state information corresponding to the current Q value, a′ represents an action corresponding to the Q value of the next state information, A represents the set of all actions corresponding to the Q value of the next state information, and θ i - Represents the parameters of the target deep Q-network model.

[0079] The mean square error loss function L of the error between the current Q value Q1 and the target Q value Q2 i The calculation formula is:

[0080] L i (θ i )=E s,a,r,s′ [(Q2-Q1) 2 ];

[0081] In this application, the gradient descent method gradually adjusts the network parameters by calculating the gradient of the mean square error loss function to reduce the value of the loss function. In the training process of the deep Q network, the gradient descent method helps to find the parameter configuration that minimizes the Q value error, so that the model can make decisions closer to the optimal under different states. By updating the parameters of the current deep Q network model by the gradient descent method and training the model in a way that minimizes the prediction error, the optimal Q value function can be effectively approached. In theory, this method gradually optimizes the network parameters through the recursive relationship of the Bellman equation, so that the model eventually learns the optimal strategy or a near-optimal strategy, ensuring that the model performs well in actual decision-making tasks. This method has been widely verified as an effective means of training deep reinforcement learning models, so the trained model is considered to be optimal or near-optimal.

[0082] The parameter update methods of the target deep Q network model include: hard update and soft update.

[0083] The hard update method is: the parameter θ of the target deep Q network model - The parameters θ of the current deep Q network model are synchronized every fixed number of steps (for example, every 1000 iterations). This ensures that the target Q value estimate of the target deep Q network model remains stable over a period of time and is not updated frequently, thereby preventing excessive fluctuations.

[0084] The soft update method is: the parameters of the target deep Q network model gradually approach the parameters of the current deep Q network model with a smaller step size. This method is achieved through the following formula:

[0085] θ - ←τθ+(1-τ)θ - ;

[0086] Where: τ is the soft update coefficient, which is usually very small, such as 0.001, which can make the update speed of the target deep Q network model slow and stable.

[0087] The target deep Q network model ensures the stability of the target Q value through a hard update or soft update strategy, thereby significantly improving the convergence and stability of the deep reinforcement learning training process.

[0088] In another exemplary embodiment of the present application, in order to further improve the accuracy of dispatching decisions, the road damage information and the geographical location information of the road image are input into a deep Q network model, a dispatching decision strategy is generated and dispatching is performed, and then the following steps are further included:

[0089] Obtain the actual maintenance effect after dispatching orders according to the dispatch decision strategy.

[0090] The reward is updated according to the actual maintenance effect, and a new state-action-reward-new state quadruple is generated.

[0091] The new state-action-reward-new state quadruple data is stored in the experience replay buffer, and the trained deep Q network model is incrementally trained based on the new state-action-reward-new state quadruple data. Incremental training refers to improving the performance of the model by continuously learning new data or experience based on the existing deep learning model, rather than training a new model from scratch. Incremental training allows the model to not only retain previously learned knowledge, but also adapt to new information. Therefore, the model can continue to improve in a dynamically changing environment. Through incremental training, deep learning models can continuously absorb new knowledge in practice and dynamically adjust decisions to achieve better performance.

[0092] Incremental training has the following advantages: ① Improve model adaptability, and adjust decision-making strategies based on the latest feedback to make better dispatch decisions. ② Avoid complete retraining, and only need to fine-tune the model with new data, saving training time and computing resources. At the same time, by giving priority to retaining the experience that has the greatest impact on the decision-making strategy, the learning efficiency of the model is improved. ③ Achieve continuous learning: Allow the system to continue learning while operating, so that the intelligence level of the system improves over time. Through incremental training, deep learning models can continuously absorb new knowledge in practice and dynamically adjust decisions to achieve better performance.

[0093] The present application also provides an application scenario, which applies the above-mentioned dispatch maintenance method for smart patrol and defect repair. Specifically: The dispatch maintenance method for smart patrol and defect repair provided in this embodiment can be applied in the urban road maintenance and management scenario. The urban road maintenance and management scenario includes links such as defect inspection, defect identification and evaluation, dispatch maintenance and effect evaluation; the process starts from the defect inspection link, and collects road defect information through vehicle-mounted equipment such as vehicle-mounted cameras. After the information is analyzed and processed in the defect identification and evaluation link, it enters the dispatch maintenance link. In the dispatch maintenance link, the dispatch maintenance method provided in this embodiment is used to automatically generate and optimize dispatch decisions based on factors such as the severity, location and urgency of road defects, and assign maintenance tasks to corresponding maintenance teams. After the maintenance team performs the maintenance task, it enters the effect evaluation link to monitor and evaluate the maintenance effect to verify the effectiveness and accuracy of the dispatch maintenance method.

[0094] The dispatch maintenance method for intelligent inspection and repair of defects provided in this embodiment belongs to the dispatch maintenance link in the urban road maintenance and management process. Specifically, in the process of road defect repair, dispatch decision optimization can be performed based on the collected road defect information and geographic location information based on the deep Q network model, that is, to automatically and efficiently generate and optimize dispatch maintenance tasks, thereby improving the response speed and maintenance quality of road defect repair and ensuring the safety and smoothness of urban roads.

[0095] Based on the same inventive concept, the embodiment of the present application also provides a maintenance system for dispatching orders for smart inspection and repair of defects for realizing the above-mentioned maintenance method for dispatching orders for smart inspection and repair of defects. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the maintenance system for dispatching orders for smart inspection and repair of defects provided below can be referred to the limitations of the maintenance method for dispatching orders for smart inspection and repair of defects above, and will not be repeated here.

[0096] In an exemplary embodiment, Figure 4 As shown, a maintenance system for intelligent inspection and repair of defects is provided, including:

[0097] The vehicle-mounted wide-angle camera 2 is installed outside the vehicle and connected to the central industrial control computer 1 to collect road images of the current road section;

[0098] A global navigation satellite system positioning antenna 3, mounted on the vehicle-mounted wide-angle camera 2, connected to the central industrial control computer 1, for obtaining geographic location information of the road image;

[0099] The central industrial control computer 1 is installed in the vehicle and includes a network module and a server. It is used to generate a dispatch decision strategy and dispatch orders based on the road image of the current road section captured by the vehicle-mounted wide-angle camera 2 and the geographical location information of the road image acquired by the global navigation satellite system positioning antenna 3, using the above-mentioned dispatch maintenance method for intelligent inspection and disease repair.

[0100] As an optional implementation, the dispatching maintenance system for intelligent inspection and repair of defects also includes:

[0101] The rotary encoder 4 is installed on the wheel and connected to the central industrial computer 1. It is used to send the pulse signal generated by the rotation of the wheel during its forward movement to the central industrial computer 1; the central industrial computer 1 calculates the distance the vehicle has moved forward based on the pulse signal, and each time the vehicle moves forward a preset distance, the central industrial computer 1 sends a command to the vehicle-mounted wide-angle camera 2 to capture an image.

[0102] Among them, a control relationship diagram and a working schematic diagram of a maintenance system for intelligent inspection and repair of defects provided in an embodiment of the present application are respectively as follows: Figure 3 and Figure 4 The vehicle-mounted wide-angle camera 2 and the Global Navigation Satellite System (GNSS) positioning antenna are shown in FIG. Figure 4 As shown, they are mounted together, wherein:

[0103] (1) The vehicle-mounted wide-angle camera 2 is powered by the central industrial control computer 1 or by the vehicle. The camera has a pixel of ≥ 2 million pixels and a clear image quality when the vehicle is in motion. The camera is installed outside the vehicle and can be used to collect road images. The collected images can clearly distinguish road manhole covers and completely capture the entire road. The image acquisition frequency can be adjusted according to the vehicle speed or controlled by the acquisition signal triggered by the central industrial control computer 1.

[0104] (2) The central industrial control computer 1 is installed in the vehicle and powered by the vehicle. The central industrial control computer 1 has a network module, such as a 4G / 5G mobile Internet module, and a server. The collected road image of the current road section and the geographical location information of the road image, as well as the road disease information in the road image obtained by inputting the road image into the deep convolutional neural network cluster, can be uploaded to the cloud server.

[0105] (3) The central industrial control computer 1 also includes a GPU processing unit, and the neural network program can use the GPU processing unit to achieve prediction acceleration;

[0106] (4) There are two forms of adjusting the frequency of image acquisition: one is to obtain the real-time speed of the vehicle through the on-board wide-angle camera 2 and the global navigation satellite system positioning antenna 3, and dynamically adjust the frequency of image acquisition according to the speed; the other is that the central industrial computer 1 is connected to the rotary encoder 4 installed on the wheel, and the central industrial computer 1 calculates the distance traveled by the vehicle based on the pulse signal generated by the rotary encoder 4. Every time the vehicle travels a preset distance, the central industrial computer 1 sends a command to the on-board wide-angle camera 2 to capture an image.

[0107] An embodiment of the present application provides a detailed implementation flow diagram of a maintenance method for dispatching orders for intelligent inspection and repair of defects, such as Figure 5 This process covers the entire process from equipment installation, image acquisition, disease detection to generation of dispatch decision strategies, dispatch, maintenance and feedback, thus realizing the automated detection and intelligent management of road diseases.

[0108] In summary, the dispatch maintenance method, system, equipment, medium and product for intelligent inspection and repair of road damage in this application, by adopting lightweight inspection equipment and deep reinforcement learning algorithm, realizes efficient collection, automatic preprocessing and intelligent dispatch decision of road damage data. The specific advantages are as follows:

[0109] Improved efficiency: Automated inspections and dispatching greatly shorten the disease treatment cycle and improve road maintenance efficiency.

[0110] Accurate decision-making: Deep reinforcement learning algorithms can simulate and optimize human decision-making habits, reduce false positives and negatives, and improve the accuracy of dispatch decisions.

[0111] Resource optimization: Through intelligent and precise dispatching, we can optimize the allocation of maintenance resources and reduce maintenance costs.

[0112] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store video tag processing data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a dispatching maintenance method for intelligent patrol and disease repair is implemented.

[0113] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0114] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0115] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0116] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0117] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0118] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0119] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0120] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0121] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A maintenance method for dispatching orders for intelligent inspection and repair of defects, characterized in that: The maintenance method for dispatching orders for intelligent inspection and repair of defects includes: Collecting a road image of the current road section and obtaining geographic location information of the road image; Inputting the road image into a deep convolutional neural network cluster to obtain road disease information in the road image, including the type of disease, the severity level of the disease, and the pixel coordinates of the disease; Whether maintenance is required is determined based on the severity of the road damage information. If maintenance is required, the road damage information and the geographic location information of the road image are input into the trained deep Q network model to generate a dispatch decision strategy and perform dispatch.

2. The maintenance method for dispatching orders for intelligent inspection and repair of defects according to claim 1 is characterized in that: Inputting the road damage information and the geographical location information of the road image into the trained deep Q network model, generating a dispatch decision strategy and performing dispatch, specifically including: Inputting state information into the trained deep Q network model to calculate the Q values ​​of all actions; the state information includes road disease information and geographical location information of the road image; The Q values ​​of all actions are compared and the action corresponding to the largest Q value is selected as the optimal action to generate a dispatch decision strategy, which includes the maintenance unit to which it belongs, the dispatch time and the priority.

3. The maintenance method for dispatching orders for intelligent inspection and repair of defects according to claim 2 is characterized in that: The step of inputting the road damage information and the geographical location information of the road image into the trained deep Q network model, generating a dispatch decision strategy and dispatching orders, further comprises: Collect historical dispatch data, including status information, dispatch decision strategy and actual maintenance effect. The actual maintenance effect includes: repair progress, repair quality, repair time and feedback information. Pre-process the historical dispatch data to generate state-action-reward-new state four-tuple data; the state is the state information, the action is the dispatch decision strategy, the reward is the reward value determined according to the actual maintenance effect, and the new state is the state information updated based on the feedback information; Constructing a deep Q network model, wherein the deep Q network model includes a current deep Q network model and a target deep Q network model; the current deep Q network model is used to generate current Q values ​​of all actions according to the input state information, and select the action corresponding to the maximum Q value as the dispatch decision strategy output; Establish an experience replay buffer to store the state-action-reward-new state quadruple data; During the training process, randomly extracting state-action-reward-new state quadruple data from the experience replay buffer to train the current deep Q network model; Acquire the network structure of the current deep Q network, construct a target deep Q network having the network structure of the current deep Q network, the target deep Q network, acquire the network parameters of the current deep Q network at fixed time intervals, and update the network parameters of the target deep Q network in a delayed update manner; The error between the current deep Q network model and the target deep Q network model is calculated using the mean square error loss function, and the parameters of the current deep Q network model are updated by minimizing the mean square error loss function through the gradient descent method to obtain the trained deep Q network model.

4. The maintenance method for dispatching orders for intelligent inspection and repair of defects according to claim 3 is characterized in that: The expression of the target Q value generated by the target deep Q network model is: Where Q2 represents the target Q value, r represents the immediate reward, γ represents the discount factor, s′ represents the next state information of the current state information corresponding to the current Q value, a′ represents an action corresponding to the Q value of the next state information, A represents the set of all actions corresponding to the Q value of the next state information, and θ i - Represents the parameters of the target deep Q-network model.

5. The maintenance method for dispatching orders for intelligent inspection and repair of defects according to claim 2 is characterized in that: The step of inputting the road damage information and the geographical location information of the road image into the trained deep Q network model, generating a dispatch decision strategy and dispatching orders, further includes: Obtain the actual maintenance effect after dispatching orders according to the dispatch decision strategy; Update the reward according to the actual maintenance effect, and generate a new state-action-reward-new state quadruple; The new state-action-reward-new-state quadruple data is stored in the experience replay buffer, and the trained deep Q network model is incrementally trained based on the new state-action-reward-new-state quadruple data.

6. A maintenance system for intelligent inspection and repair of defects, characterized in that: The dispatching maintenance system for intelligent inspection and repair of defects includes: The vehicle-mounted wide-angle camera is installed outside the vehicle and connected to the central industrial computer to collect road images of the current section; A global navigation satellite system positioning antenna, mounted on the vehicle-mounted wide-angle camera and connected to a central industrial control computer, for obtaining geographic location information of the road image; A central industrial control computer is installed in the vehicle and includes a network module and a server. It is used to generate a dispatch decision strategy and perform dispatch according to the road image of the current road section captured by the vehicle-mounted wide-angle camera and the geographic location information of the road image acquired by the global navigation satellite system positioning antenna, using the dispatch maintenance method for intelligent patrol inspection and repair as described in claims 1-5.

7. The maintenance system for intelligent inspection and repair of defects according to claim 6 is characterized in that: The dispatching maintenance system for intelligent inspection and repair of defects also includes: A rotary encoder is installed on the wheel and connected to the central industrial computer. It is used to send a pulse signal generated by the rotation of the wheel during its forward movement to the central industrial computer. The central industrial computer calculates the distance the vehicle has traveled based on the pulse signal. Every time the vehicle travels a preset distance, the central industrial computer sends a command to the on-board wide-angle camera to capture an image.

8. A computer device comprising: 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 dispatching maintenance method for intelligent inspection and defect repair as described in any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the dispatching maintenance method for intelligent inspection and repair of defects as described in any one of claims 1-5.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the dispatching maintenance method for intelligent inspection and repair of defects as described in any one of claims 1-5.

Images