A full life cycle closed loop management method and system for highway maintenance

Abstract

The application relates to the technical field of intelligent management of highway maintenance, and particularly discloses a full-life-cycle closed-loop management method and system for highway maintenance, which collects multi-source highway disease data through a multi-source sensing device, constructs a disease confidence analysis mechanism, and evaluates the accuracy of highway disease identification; based on a reinforcement learning algorithm, optimal maintenance dispatching strategies are dynamically generated by comprehensively considering disease severity and maintenance constraints, electronic work orders are generated and pushed to maintenance construction units; a highway image registration algorithm is used to align highway images before and after maintenance to capture repair evaluation indexes, a multi-dimensional scoring algorithm is used to calculate maintenance quality scores, and acceptance levels are determined; the maintenance quality scores are used as feedback signals to drive reinforcement learning model parameter updating, automatically optimize dispatching rules and disposal strategies, and construct a disease life cycle atlas. The application improves the accuracy and timeliness of disease identification, and strengthens the highway operation safety and the utilization efficiency of maintenance resources.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for highway maintenance, and more specifically, to a closed-loop management method and system for the entire life cycle of highway maintenance. Background Technology

[0002] With the continuous growth of highway mileage and traffic flow in my country, the importance of highway maintenance has become increasingly prominent. Timely detection, effective treatment, and closed-loop management of highway defects are key aspects of ensuring safe and smooth road traffic. However, existing technologies face several pain points in practical applications: First, the processes of disease detection and treatment are disconnected. In traditional maintenance management models, there is a lack of effective data correlation between disease detection, task assignment, construction implementation, and acceptance evaluation, resulting in a discontinuous maintenance management chain and difficulty in forming a complete closed loop. Second, existing acceptance methods mainly rely on manual on-site inspections or review of construction photos, which suffers from low efficiency, strong subjectivity, and inconsistent evaluation standards, making it impossible to objectively and quantitatively assess the road repair effect. At the same time, existing maintenance task assignment methods mostly use fixed rules, such as allocation based on distance priority or work order order, without fully considering dynamic factors such as the severity of the disease, the capabilities of the construction unit, the road operating status, and historical construction results, leading to insufficient efficiency in the allocation of maintenance resources. In addition, existing systems generally lack feedback and learning mechanisms based on historical maintenance data, making it impossible to continuously optimize task allocation and maintenance strategies based on actual treatment results, which can easily lead to the recurrence of similar diseases and waste of maintenance resources.

[0003] Therefore, it is necessary to provide a closed-loop management method and system for the entire life cycle of highway maintenance to solve the above-mentioned technical problems. In order to solve the above problems, a technical solution is provided. Summary of the Invention

[0004] To overcome the aforementioned deficiencies in the prior art, this invention provides a closed-loop management method and system for the entire life cycle of highway maintenance, which addresses the problems of disconnect between disease detection and treatment, subjective and inefficient acceptance, rigid task allocation, and lack of data feedback loop in current highway maintenance.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A closed-loop management method for the entire life cycle of highway maintenance includes the following steps: Multi-source highway damage data are collected through multi-source sensing devices, and image recognition algorithms are used to automatically identify the type, location and severity of highway damage, generating standardized damage records for the damage database. A road defect confidence analysis mechanism was constructed based on multi-source road defect data to evaluate the accuracy of road defect identification. Based on reinforcement learning algorithms, the optimal maintenance dispatch strategy is dynamically generated by considering the severity of the disease and maintenance constraints, and electronic work orders are generated and pushed to the maintenance construction unit. Real-time tracking of the maintenance and construction process, recording on-site attendance, progress reports and construction data, and supporting abnormal warnings and dynamic secondary dispatching; A highway image registration algorithm is used to align the highway image capture and repair evaluation indicators before and after maintenance. A multi-dimensional scoring algorithm is used to calculate the maintenance quality score and determine the acceptance level. The maintenance quality score is used as a feedback signal to drive the update of reinforcement learning model parameters and automatically optimize dispatch rules and handling strategies. By constructing a disease life cycle map, the evolution trajectory of diseases at the same location and the effects of previous treatments are recorded, enabling preventive maintenance decisions.

[0006] As a further aspect of the present invention, a road defect confidence analysis mechanism is constructed based on multi-source road defect data to evaluate the accuracy of road defect identification. The specific steps are as follows: A multi-source highway defect data was acquired to establish a defect confidence analysis mechanism and calculate confidence analysis indicators. A weighted model is constructed based on confidence analysis indicators to obtain the comprehensive road damage confidence level. The comprehensive road damage confidence level is then compared with a preset confidence threshold to determine whether the confidence level of the corresponding road damage data is acceptable. If the comprehensive road damage confidence level is greater than or equal to the preset confidence threshold, the confidence level of the corresponding road damage data is acceptable; if the comprehensive road damage confidence level is less than the preset confidence threshold, the confidence level of the corresponding road damage data is unacceptable. The accuracy of highway defect identification is evaluated based on the confidence level of highway defect data. The accuracy of highway defect identification is the ratio of the number of highway defect data with qualified confidence to the total number of highway defect data.

[0007] As a further aspect of the present invention, the confidence analysis indicators include image recognition confidence, spatial consistency confidence, and temporal continuity confidence.

[0008] As a further aspect of the present invention, based on a reinforcement learning algorithm, considering both the severity of the disease and maintenance constraints, an optimal maintenance dispatch strategy is dynamically generated, and an electronic work order is generated and pushed to the maintenance construction unit. The specific steps are as follows: Obtain the severity of the disease and maintenance constraints; maintenance constraints include the conditions of the maintenance construction unit and the environmental conditions. Based on reinforcement learning algorithms, maintenance task actions are constructed according to the current highway status; By setting up a multi-dimensional maintenance task assignment reward function, the merits and demerits of maintenance task actions can be evaluated from multiple dimensions.

[0009] As a further aspect of the present invention, the status constraints of the maintenance and construction unit include the current number of maintenance and construction tasks, the level of maintenance and construction technical capabilities, historical acceptance scores, and distance from the location of the defects; the environmental status constraints include weather conditions, road traffic flow, and the available construction time window.

[0010] As a further aspect of the present invention, the multi-dimensional maintenance dispatch reward function includes rewards for timely treatment of defects, rewards for construction quality, rewards for cost optimization, and rewards for traffic impact.

[0011] As a further aspect of the present invention, a highway image registration algorithm is used to align the highway image capture and repair evaluation indicators before and after maintenance, and a multi-dimensional scoring algorithm is used to calculate the maintenance quality score and determine the acceptance level. The specific steps are as follows: Before and after highway maintenance images were acquired, and image preprocessing and alignment operations were performed, including image size normalization, image enhancement, and extraction of diseased areas. Repair assessment indicators were captured from aligned pre-maintenance and post-maintenance highway images. The maintenance quality score is calculated based on the repair assessment indicators input into a multi-dimensional scoring algorithm to determine the acceptance level. The maintenance quality score is compared with a preset scoring range to determine the acceptance level. When the maintenance quality score is greater than the upper limit of the preset scoring range, it is at the first-level acceptance level and automatically passes the acceptance. When the maintenance quality score is within the preset scoring range, it is at the second-level acceptance level and requires manual review. When the maintenance quality score is less than the lower limit of the preset scoring range, it is at the third-level acceptance level and triggers the rework process.

[0012] As a further aspect of the present invention, the repair evaluation indicators include area repair rate, texture similarity, edge clarity, and color difference.

[0013] A closed-loop management system for the entire life cycle of highway maintenance includes a multi-source sensing module, a data confidence analysis module, an intelligent dispatching module, a construction tracking module, a maintenance quality assessment module, a strategy optimization engine, and a map management module. The multi-source sensing module collects multi-source highway disease data through multi-source sensing devices, and uses image recognition algorithms to automatically identify the type, location and severity of highway diseases, and generates standardized disease records to the disease database. The data confidence analysis module is used to construct a road defect confidence analysis mechanism based on multi-source road defect data and to evaluate the accuracy of road defect identification. The intelligent work order dispatching module is used to dynamically generate the optimal maintenance work order dispatching strategy based on reinforcement learning algorithms, taking into account the severity of the disease and maintenance constraints, and to generate and push electronic work orders to the maintenance construction unit. The construction tracking module is used to track the maintenance construction process in real time, record attendance, progress reports and construction data, and support abnormal warnings and dynamic secondary dispatching. The maintenance quality assessment module is used to align the highway image capture and repair assessment indicators before and after maintenance using a highway image registration algorithm, and to calculate the maintenance quality score and determine the acceptance level through a multi-dimensional scoring algorithm. The strategy optimization engine uses maintenance quality scores as feedback signals to drive the update of reinforcement learning model parameters and automatically optimize dispatch rules and handling strategies. The map management module is used to construct a disease life cycle map, record the evolution trajectory of diseases at the same location and the effects of previous treatments, and make preventive maintenance decisions.

[0014] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of a closed-loop management method and system for the entire life cycle of highway maintenance as described in any of the above claims.

[0015] The technical effects and advantages of this invention, a closed-loop management method and system for the entire lifecycle of highway maintenance, are as follows: This invention achieves closed-loop management of highway defects from automatic detection, intelligent analysis, task scheduling, construction supervision, quality acceptance to strategy optimization. Compared with traditional maintenance methods that rely on manual inspection and fixed-rule dispatching, this invention utilizes multi-source data fusion technology to improve the accuracy and timeliness of defect identification, reduces misidentification and missed identification problems in complex environments through a confidence evaluation mechanism, and dynamically adjusts maintenance dispatching strategies based on defect severity, construction unit capabilities, and environmental conditions using reinforcement learning algorithms, thereby improving the efficiency of maintenance resource allocation and task response speed.

[0016] This invention achieves objective and quantitative evaluation of maintenance effectiveness through AI image registration and multi-dimensional quality scoring algorithms, reducing subjective errors caused by manual acceptance and improving the consistency of acceptance standards. Real-time monitoring of the construction process and an anomaly early warning mechanism can promptly detect deviations in task execution and make dynamic adjustments. Furthermore, analysis of historical disease evolution patterns based on disease lifecycle mapping enables a shift from passive maintenance to proactive predictive maintenance, reducing repetitive maintenance costs and improving highway operational safety and the level of intelligent maintenance management. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a closed-loop management method for the entire lifecycle of highway maintenance, provided as an embodiment of the present invention; Figure 2 This is a system block diagram of a closed-loop management system for the entire life cycle of highway maintenance, provided as an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described technical solutions are only a part of this invention, and not all of it. All other technical solutions obtained by those skilled in the art based on the technical solutions of this invention without inventive effort are within the scope of protection of this invention.

[0019] Example 1: As Figure 1 The diagram shown is a flowchart of a closed-loop management method for the entire life cycle of highway maintenance provided by an embodiment of the present invention. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S7 are detailed below: Step S1: Collect multi-source highway defect data through multi-source sensing devices, and automatically identify the type, location, and severity of highway defects using image recognition algorithms to generate standardized defect records for the defect database; the multi-source highway defect data includes highway images, highway video information, GPS spatial coordinates, and data collection timestamps; the standardized defect records include fields such as station number, lane, defect type, severity level, first discovery time, and on-site images; This invention utilizes multi-source sensing devices, such as vehicle-mounted inspection equipment, drone inspection equipment, and fixed road monitoring equipment, to continuously collect data on target highway areas, acquiring multi-source highway defect data including highway images, highway video information, GPS spatial coordinates, and data acquisition timestamps. Specifically, highway images and video information describe the road surface condition, GPS spatial coordinates determine the spatial location of defects, and data acquisition timestamps record the time of defect discovery and subsequent evolution analysis.

[0020] The collected road images and video data are input into a deep learning-based image recognition model. A convolutional neural network extracts road surface texture, morphology, and edge features to automatically identify types of road defects such as cracks, potholes, subsidence, bulges, and network cracks. Combined with target detection algorithms, the system determines the location, affected area, and severity of the defects. Based on the identification results, the system generates standardized defect records and stores them in a defect database. These standardized defect records include fields such as road station number, lane number, defect type, severity level, first discovery time, and on-site image capture, enabling structured management of defect information and providing a data foundation for subsequent defect reliability analysis, intelligent dispatching, and lifecycle management.

[0021] Step S2: Construct a road defect confidence analysis mechanism based on multi-source road defect data to evaluate the accuracy of road defect identification; After completing the identification of road defects, this invention constructs a defect confidence analysis mechanism based on multi-source highway defect data to evaluate the credibility of the automatically identified defect results. Specifically, it comprehensively utilizes image recognition results, spatial positioning information, and historical detection data to construct a multi-dimensional evaluation index, including image recognition confidence, spatial consistency confidence, and temporal continuity confidence.

[0022] The image recognition confidence level is obtained by analyzing the disease category probabilities output by the deep learning model; the spatial consistency confidence level is calculated by comparing the positional deviation between the AI-identified disease location coordinates and the road coordinates on the high-precision map; and the temporal continuity confidence level is calculated by analyzing the trend of disease area changes at different time points in the same area. Subsequently, the various confidence level indicators are weighted and fused to obtain the comprehensive disease confidence level, which is then compared with a preset confidence threshold. When the comprehensive confidence level reaches the threshold requirement, the corresponding disease data is determined to be reliable data; otherwise, it is marked as data to be reviewed.

[0023] The accuracy rate of highway defect identification is calculated based on the ratio between the number of defect data with acceptable confidence level and the total number of identified defects. This allows for a quantitative evaluation of the actual performance of the defect identification model and improves the reliability of defect detection results.

[0024] Step S3: Based on reinforcement learning algorithm, the optimal maintenance dispatch strategy is dynamically generated by comprehensively considering the severity of the disease and maintenance constraints, and an electronic work order is generated and pushed to the maintenance construction unit. This invention acquires disease data that has undergone credibility verification, and constructs a reinforcement learning state space based on the severity of the disease and maintenance constraints. The maintenance constraints include state constraints for the maintenance construction unit and environmental constraints. Specifically, the state constraints for the maintenance construction unit include the current number of construction tasks, technical capability level, historical acceptance scores, and distance from the disease location; the environmental constraints include weather conditions, road traffic flow, and a feasible construction time window.

[0025] Based on the current state of highway defects, a reinforcement learning model constructs multiple maintenance tasks, including options for selecting different construction companies, determining task execution times, and allocating construction resources. The model comprehensively evaluates different dispatch actions by setting rewards for timeliness of defect treatment, construction quality, cost optimization, and traffic impact, and continuously optimizes the action selection strategy based on the reward values.

[0026] The reinforcement learning model selects the maintenance task action with the highest overall benefit, determines the optimal construction unit, construction time, and resource allocation plan, and automatically generates an electronic maintenance work order. The electronic work order includes the defect number, defect location, defect type, severity level, repair requirements, construction period, and construction unit information, and is pushed to the corresponding construction unit through the maintenance management platform to achieve automated allocation of maintenance tasks.

[0027] Step S4: Track the maintenance and construction process in real time, record attendance, progress reports and construction data, and support abnormal warnings and dynamic reassignment. In this embodiment of the invention, after the electronic work order is pushed to the maintenance and construction unit, the construction process monitoring stage begins. Construction execution data is acquired in real time through the construction unit's mobile terminal, on-site data collection equipment, and the maintenance management platform. After receiving the work order, construction personnel confirm the task via their mobile terminal and upload their attendance information. GPS location data is used to determine whether the construction personnel have arrived at the designated defect location as required.

[0028] During construction, the construction unit continuously uploads construction progress information, including the start time, current completion percentage, site photos, material usage, and equipment deployment, enabling dynamic tracking of the maintenance process. When abnormal construction progress, failure to arrive on time, construction quality risks, or changes in environmental conditions are detected, an automatic warning mechanism is triggered. For maintenance tasks that cannot be completed as planned, the reinforcement learning-based task assignment model is dynamically optimized based on the current construction status and real-time resource availability. This generates a secondary task assignment strategy, reallocating the task to a suitable construction unit, thus achieving dynamic adjustments during maintenance task execution and improving the flexibility of maintenance management.

[0029] Step S5: Use a highway image registration algorithm to align the highway images before and after maintenance to capture and evaluate the repair indicators. Calculate the maintenance quality score using a multi-dimensional scoring algorithm to determine the acceptance level. The repair evaluation indicators include area repair rate, texture similarity, edge clarity, and color difference. After maintenance work was completed, images of the highway before and after maintenance were acquired, and image correspondences were established based on road markers, GPS coordinates, and work order information. First, the images were normalized, enhanced, and damaged areas were extracted to reduce the impact of factors such as shooting angle and lighting changes on the evaluation results. Then, a highway image registration algorithm was used to extract feature points from the images before and after maintenance, and image spatial correspondences were established through feature descriptor matching to achieve precise alignment of the maintenance areas. After image registration, restoration evaluation indicators such as area restoration rate, texture similarity, edge sharpness, and color difference were extracted. The area restoration rate was used to evaluate the degree of damage area elimination; texture similarity was used to evaluate the consistency of texture between the restored area and the original road surface; edge sharpness was used to evaluate the smoothness of the construction boundaries; and color difference was used to evaluate the visual fusion effect of the restored area.

[0030] The above indicators are input into a multi-dimensional scoring algorithm to calculate a comprehensive maintenance quality score, and the acceptance level is divided according to a preset scoring range. When the score reaches the first-level acceptance standard, the acceptance is automatically passed; when the score is in the middle range, it enters manual review; when the score is below the standard, the rework process is triggered, realizing intelligent evaluation of maintenance quality.

[0031] Step S6: Use the maintenance quality score as a feedback signal to drive the update of reinforcement learning model parameters and automatically optimize dispatch rules and handling strategies. This invention uses maintenance quality scores as feedback signals for a reinforcement learning model to evaluate the effectiveness of historical work order allocation strategies. When a maintenance work order allocation strategy has a high maintenance quality score and simultaneously meets conditions such as fast construction response, low cost, and minimal traffic impact, the value of the corresponding action is increased. Conversely, when a strategy leads to lower repair quality, increased rework, or significant construction impact, the value of the corresponding action is decreased. The reward function parameters and action value function in the reinforcement learning model are adjusted based on the maintenance quality scores, enabling the model to continuously learn from historical maintenance task execution results. By continuously accumulating the correlations between disease types, construction units, work order allocation schemes, and acceptance results, subsequent maintenance task allocation strategies are optimized.

[0032] After multiple rounds of task execution and feedback updates, the reinforcement learning model can automatically select more reasonable maintenance plans for different types of defects, different road environments, and different construction resource conditions, thereby improving the accuracy of maintenance dispatch and overall management efficiency.

[0033] Step S7: By constructing a disease life cycle map, the evolution trajectory of diseases at the same location and the effects of previous treatments are recorded to make preventive maintenance decisions.

[0034] Based on historically collected data on road defects, maintenance processes, and acceptance evaluations, a road defect lifecycle map is constructed to describe the entire process of defects at the same road location, from initial discovery, development, repair, and subsequent restoration. Using road spatial locations as correlation nodes, the system links defect types, severity, area change trends, repair plans, construction units, and acceptance results from each detection, forming a defect evolution trajectory. For example, for problem areas with repeated cracks or potholes on the same road section, the system analyzes historical defect growth patterns to determine their potential risk level.

[0035] Based on the disease lifecycle map, and further combined with disease development trends and historical maintenance results, the location and type of potential future diseases can be predicted, and preventative maintenance plans can be developed in advance, including early inspections, localized repairs, and specialized maintenance measures. Through the accumulation of lifecycle data, a proactive maintenance model can be shifted from the traditional "repair after disease is discovered" to "predict disease and intervene in advance," thereby improving highway operation safety and the efficiency of maintenance resource utilization.

[0036] Preferably, an image recognition algorithm is used to automatically identify the type, location, and severity of highway defects; a deep learning image recognition algorithm (based on a convolutional neural network CNN) is used to analyze the collected highway images in real time, automatically identify common highway defect types such as cracks, potholes, subsidence, bulges, and network cracks, and mark the location, area, and severity of the highway defects.

[0037] Preferably, a road defect confidence analysis mechanism is constructed based on multi-source road defect data to evaluate the accuracy of road defect identification. The specific steps are as follows: A road damage confidence analysis mechanism was established by acquiring multi-source road damage data, and confidence analysis indicators were calculated. The confidence analysis indicators include image recognition confidence, spatial consistency confidence, and temporal continuity confidence. A weighted model is constructed based on confidence analysis indicators to obtain the comprehensive road damage confidence level. The comprehensive road damage confidence level is then compared with a preset confidence threshold to determine whether the confidence level of the corresponding road damage data is acceptable. If the comprehensive road damage confidence level is greater than or equal to the preset confidence threshold, the confidence level of the corresponding road damage data is acceptable; if the comprehensive road damage confidence level is less than the preset confidence threshold, the confidence level of the corresponding road damage data is unacceptable. The accuracy of highway defect identification is evaluated based on the confidence level of highway defect data. The accuracy of highway defect identification is the ratio of the number of highway defect data with qualified confidence to the total number of highway defect data.

[0038] Preferably, the confidence analysis indicators include image recognition confidence, spatial consistency confidence, and temporal continuity confidence, as detailed below: The specific details of image recognition confidence are: Using a CNN model to output the maximum classification probability for image recognition ,in, Number of disease categories This refers to the maximum classification probability in image recognition. Normalization is performed to obtain the image recognition confidence score. .

[0039] The spatial consistency confidence is specifically calculated by matching the location coordinates obtained from the highway image (AI recognition location coordinates) and the high-precision map location coordinates. ; In the formula: This represents the image position deviation value. To enable AI to identify location coordinates, The location coordinates are for a high-precision map. Calculate spatial consistency confidence based on image positional deviation values: ; In the formula: For spatial consistency confidence. The spatial attenuation coefficient, This represents the area of ​​road damage detected at the previous moment. Calculate the time continuity confidence level based on the rate of change of highway defects detected: ; In the formula: For time continuity confidence, Weights are determined by time variation.

[0040] In this embodiment, by deploying vehicle-mounted inspection equipment, drone inspection equipment, and fixed road monitoring equipment, continuous data collection is carried out on the target highway section to obtain multi-source highway defect data containing road images, spatial location, and historical detection information. Based on deep learning image recognition algorithms, automatic identification of highway defects is achieved. At the same time, a defect confidence analysis mechanism is combined to evaluate the credibility of the identification results, thereby improving the accuracy and reliability of defect identification results.

[0041] Vehicle-mounted inspection equipment continuously collects images of target highway sections according to a preset inspection route, acquiring high-definition road image data containing road surface condition information. Simultaneously, drone inspection equipment supplements the collection of images from key areas, complex structural areas, and areas difficult to cover manually, obtaining road image data from different perspectives. In addition, fixed road cameras acquire long-term road monitoring video data, and simultaneously record the GPS spatial coordinates, collection time, and corresponding road marker information for each data collection process, thus forming a multi-source highway defect data set that includes image data, spatial location data, and temporal change data.

[0042] After acquiring highway image data, a deep learning image recognition algorithm based on convolutional neural networks (CNNs) is used to analyze the acquired road images in real time. First, the input road images are preprocessed, including image resizing, noise filtering, and image enhancement, to improve image feature extraction. Then, the processed road images are input into a pre-trained CNN-based defect identification model. Convolutional layers extract road surface texture, edge, and morphological features, and a classification network outputs the probability of the corresponding defect category. Based on the model output, the defect types in the current road image are automatically identified, including common highway defect types such as cracks, potholes, subsidence, bulges, and network cracks.

[0043] The target detection algorithm is used to locate the identified diseased areas. The specific location of the diseased areas in the road image is obtained by the detection box annotation method, and the area of ​​the diseased area is calculated by combining the image pixel information. Furthermore, the severity of the disease is analyzed by combining the size of the diseased area, the scope of influence, and the degree of impact on road safety, and a structured disease identification result including disease type, disease location, disease area, and severity level is generated.

[0044] To avoid misidentification caused by factors such as changes in lighting, shooting angle, and road environment in a single image recognition result, a disease confidence analysis mechanism was further established based on the acquired multi-source highway disease data to evaluate the credibility of each disease recognition result.

[0045] Calculate the image recognition confidence score. Use a CNN model to output the predicted probability of the current road image for different disease categories. Since the probability output range varies among different image samples, the maximum classification probability is normalized and converted into an image recognition confidence score within the range of 0 to 1. When the CNN model has a high predicted probability for a certain disease category, it indicates that the model's recognition result for that disease type is more reliable, and the corresponding image recognition confidence score is high.

[0046] Calculate spatial consistency confidence. Obtain the coordinates of the lesion location identified by the CNN model, and call the high-precision map database to obtain the corresponding actual spatial location coordinates of the road. Calculate the positional deviation between the two by coordinate matching. Further calculate the spatial consistency confidence based on the positional deviation value: when the AI-identified location is closer to the road location on the high-precision map, the positional deviation value decreases, and the corresponding spatial consistency confidence increases, indicating that the lesion location result is more accurate.

[0047] Calculate the confidence level for temporal continuity. Retrieve historical inspection data for the same road area and analyze the changes in disease area detected at adjacent time points. Calculate the change in disease area between the current and previous time points. Calculate the confidence level for temporal continuity based on the rate of change in disease area. When the change in disease area conforms to a normal development trend, the rate of change in disease area is small, and the confidence level for temporal continuity is high; when abnormal abrupt changes occur in the detection results, the corresponding confidence level is reduced, and further verification is prompted.

[0048] After obtaining the confidence scores for image recognition, spatial consistency, and temporal continuity, a multi-index weighted evaluation model is established to calculate the comprehensive disease confidence score. By comprehensively considering the image recognition results, spatial positioning accuracy, and temporal continuity, a single recognition result is transformed into a credible evaluation result under the joint constraints of multi-source data.

[0049] The calculated overall disease confidence score is compared with a preset confidence threshold. ,in, If a pre-set confidence threshold is set, the current disease identification result is deemed reliable, the disease data is written into the disease database, and the process proceeds to subsequent intelligent dispatching and maintenance. If the current disease identification result is deemed unreliable, the data will be marked as data to be reviewed and further verified by re-collecting images, recalling historical data, or manual assistance.

[0050] The accuracy of disease identification is evaluated based on the confidence level of all disease identification data. This evaluation method can intuitively reflect the reliability of the current disease identification model in actual road environments, and continuously optimize the image recognition model parameters based on low-confidence samples.

[0051] Through the above implementation methods, intelligent identification and reliability evaluation based on multi-source highway defect data are realized. The image recognition results, spatial positioning information and temporal change patterns are fused and analyzed, which effectively reduces the problems of misidentification and omission in complex road environments, improves the accuracy of highway defect detection results, and provides a reliable data foundation for subsequent intelligent allocation of maintenance tasks and closed-loop management throughout the entire life cycle.

[0052] Preferably, based on a reinforcement learning algorithm, the optimal maintenance dispatch strategy is dynamically generated by considering both the severity of the damage and maintenance constraints. Electronic work orders are then generated and pushed to the maintenance contractor. The specific steps are as follows: Obtain the severity of the disease and maintenance constraints; maintenance constraints include the status constraints of the maintenance construction unit and the environmental status constraints; the status constraints of the maintenance construction unit include the current number of maintenance construction tasks, the level of maintenance construction technical capabilities, historical acceptance scores, and distance from the location of the disease; the environmental status constraints include weather conditions, road traffic flow, and the available construction time window. Based on the reinforcement learning algorithm, maintenance task actions are constructed according to the current highway state. The set of maintenance task actions is as follows: , For the j-th maintenance and construction task action, Total number of maintenance and construction tasks; ,in, For the selected maintenance and construction company, For task execution time, This is a resource allocation plan. Specifically, for example, when a high-grade pothole defect is discovered on a road section, the following maintenance tasks are triggered: Action 1: Select the nearest construction unit; Action 2: Select the unit with the highest historical repair quality; Action 3: Select the unit with the highest overall efficiency. By setting up a multi-dimensional maintenance dispatch reward function, the merits and demerits of maintenance tasks are evaluated from multiple dimensions. The multi-dimensional maintenance dispatch reward function includes rewards for timeliness of disease treatment, construction quality, cost optimization, and traffic impact.

[0053] Preferably, the timeliness reward for disease treatment is used to evaluate maintenance task actions. The faster the actual response time, the higher the output of the timeliness reward for disease treatment, as expressed by the formula:

[0054] In the formula: Rewards will be given for timely disease treatment. To specify the processing time, This represents the actual response time.

[0055] Construction quality rewards are determined based on historical AI acceptance scores; the higher the historical AI acceptance score, the greater the construction quality reward output.

[0056] The cost optimization reward takes into account the construction cost, which includes personnel input cost, equipment usage cost, and material consumption cost. The lower the construction cost, the greater the cost optimization reward output.

[0057] Traffic impact compensation is determined based on the degree of road impact during construction, using the following formula: ; In the formula: Based on the degree of impact on the highway, Traffic volume during construction period This is normal traffic volume.

[0058] In this embodiment of the invention, when road defects are detected by the front-end inspection equipment and image recognition module, maintenance tasks to be processed are generated based on the defect identification results. Based on the reinforcement learning algorithm, the severity of the defects, the status of the construction unit, and the constraints of the road environment are comprehensively analyzed to dynamically generate the optimal maintenance dispatch strategy, thereby realizing automatic allocation of maintenance tasks and intelligent push of electronic work orders.

[0059] The system acquires information on the severity of existing road defects, output by the preceding defect identification module. This information includes defect type, area, affected range, and severity level. For example, when potholes are identified on a highway section, their severity is classified based on their area, depth, lane location, and impact on traffic safety. If a large pothole is detected, located in a high-speed lane, and poses a safety risk to vehicles, the defect is marked as a high-level emergency maintenance task, and its processing priority is increased.

[0060] While obtaining the severity of the damage, the system simultaneously acquires data on the constraints in the current maintenance task execution process. These constraints include the status constraints of the maintenance construction unit and the environmental status constraints.

[0061] The status constraints of maintenance construction units primarily describe the current availability of construction resources for participating in maintenance tasks. These include the number of current maintenance tasks, the technical capability level of the construction unit, historical acceptance scores, and the distance of the construction unit from the location of the defect. The system obtains the current task load of each construction unit from the maintenance management database. When a construction unit undertakes multiple maintenance tasks simultaneously, its task allocation priority is reduced. Simultaneously, the system obtains the technical capability level and historical AI acceptance scores of construction units based on their historical construction records. For construction units with higher repair quality and lower rework rates, their weight in task allocation is increased. Furthermore, by acquiring the real-time location information of construction units, the system calculates the distance between the construction unit and the defect location, serving as a basis for evaluating task response efficiency.

[0062] Environmental constraints describe the current road maintenance implementation environment, including real-time weather conditions, road traffic flow, and feasible construction time windows. Weather conditions determine whether the current environment meets construction requirements; for example, severe weather conditions such as rain and fog may affect construction quality, thus reducing the suitability of the corresponding construction plan. Road traffic flow evaluates the impact of the construction process on road capacity; when the target road segment is under peak traffic, priority should be given to construction times with minimal traffic impact. The feasible construction time window limits the execution time of the construction task to ensure that the maintenance process meets road operation and management requirements.

[0063] After acquiring information on the condition of road defects and maintenance constraints, the system constructs a maintenance task action space based on a reinforcement learning algorithm, treating different maintenance dispatch schemes as selectable actions. Each maintenance task action represents a specific task allocation scheme. For example, when a high-grade pothole defect is detected on a highway section, multiple candidate maintenance task actions are generated based on the current road condition, including: Action 1, selecting the construction unit closest to the defect location to reduce construction response time; Action 2, selecting the construction unit with the highest historical repair quality to improve the quality of defect repair; Action 3, selecting the construction unit with the highest overall efficiency, i.e., determining the optimal unit after comprehensively considering construction distance, historical acceptance scores, task load, and construction cost. The reinforcement learning model selects the optimal maintenance task action from multiple candidate actions by predicting and evaluating the execution effects of different actions.

[0064] To achieve accurate evaluation of different maintenance tasks, a multi-dimensional maintenance dispatch reward function is further constructed to quantitatively analyze the comprehensive benefits generated by each maintenance task. This multi-dimensional maintenance dispatch reward function includes rewards for timely disease treatment, construction quality, cost optimization, and traffic impact.

[0065] The timeliness bonus for disease treatment is used to evaluate the construction unit's response efficiency to disease tasks. The timeliness bonus is calculated by comparing the specified treatment time with the actual response time. When the construction unit can respond quickly and arrive at the site promptly after receiving the work order, the actual response time is shorter, resulting in a higher calculated bonus, indicating that the maintenance task has high execution efficiency.

[0066] Construction quality rewards are used to evaluate the historical maintenance performance of different construction companies. The system retrieves AI acceptance scores from the historical maintenance database and uses the quality of the repairs after completion as the basis for rewards. When a construction company receives a high AI acceptance score in its historical maintenance tasks, it indicates superior construction quality, and the corresponding construction quality reward output is increased; conversely, for construction companies with a high number of rework cases or low acceptance scores, their construction quality rewards are decreased.

[0067] Cost optimization rewards are used to evaluate the economic efficiency of maintenance tasks. The system comprehensively considers the costs of construction personnel, equipment usage, and material consumption to calculate the costs of different construction plans. Specifically, when a construction plan can reduce the input of personnel, equipment, and materials while meeting the requirements for disease repair, its corresponding cost optimization reward is increased; when the cost of a construction plan is too high, the corresponding reward value is reduced.

[0068] Traffic impact reward is used to evaluate the degree of impact of construction on normal road operation. It is calculated based on changes in road traffic during construction. When the construction plan is carried out during periods of low traffic volume or when rapid construction methods are used to reduce road occupancy time, the traffic impact during construction is smaller, the corresponding traffic volume is lower, and the calculated degree of road impact increases, indicating that the task is more in line with road operation requirements.

[0069] Furthermore, the aforementioned multiple reward indicators are integrated to form an evaluation value for maintenance dispatch actions, and the value function corresponding to different actions is continuously updated through a reinforcement learning model. Based on the current disease status input, the model calculates the expected return of each candidate maintenance task action and selects the action with the highest overall reward as the final maintenance dispatch strategy.

[0070] Once the reinforcement learning model determines the optimal maintenance task action, an electronic maintenance work order is automatically generated based on the corresponding strategy. The electronic work order includes the defect number, defect type, severity level, defect location, on-site images, construction requirements, planned completion time, construction unit information, and resource allocation plan, and is pushed to the corresponding construction unit's mobile terminal through the maintenance management platform.

[0071] After receiving the electronic work order, the construction unit confirms the task via a mobile terminal and arrives at the site to carry out maintenance work based on the defect location information in the work order. During the construction process, the construction unit continuously uploads construction progress, site photos, material usage, and equipment deployment information to track the task execution process in real time. After the construction is completed, the repair effect is evaluated through an AI acceptance module, and the acceptance results are fed back to a reinforcement learning model to optimize subsequent maintenance dispatch strategies.

[0072] The embodiments of the present invention realize dynamic allocation of highway maintenance tasks based on reinforcement learning, which enables the maintenance task assignment process to comprehensively consider the severity of defects, the status of construction resources and road environmental constraints. Compared with the traditional fixed rule assignment method, it improves the efficiency of maintenance resource matching and the timeliness of defect treatment.

[0073] Preferably, a highway image registration algorithm is used to align the highway image capture and restoration evaluation indicators before and after maintenance. A multi-dimensional scoring algorithm is used to calculate the maintenance quality score and determine the acceptance level. The restoration evaluation indicators include area restoration rate, texture similarity, edge clarity, and color difference. The specific steps are as follows: Before and after highway maintenance images were acquired, and image preprocessing and alignment operations were performed, including image size normalization, image enhancement, and extraction of diseased areas. Repair assessment indicators were captured from aligned pre-maintenance and post-maintenance highway images. The maintenance quality score is calculated based on the repair assessment indicators input into a multi-dimensional scoring algorithm to determine the acceptance level. The maintenance quality score is compared with a preset scoring range to determine the acceptance level. When the maintenance quality score is greater than the upper limit of the preset scoring range, it is at the first-level acceptance level and automatically passes the acceptance. When the maintenance quality score is within the preset scoring range, it is at the second-level acceptance level and requires manual review. When the maintenance quality score is less than the lower limit of the preset scoring range, it is at the third-level acceptance level and triggers the rework process.

[0074] The specific formula for the multi-dimensional scoring algorithm is as follows: ; In the formula: To score the quality of maintenance, This is the weighting coefficient for the area restoration rate. For area restoration rate, These are the weighting coefficients for texture similarity. For texture similarity, This is a weighting coefficient for edge sharpness. For edge sharpness, This is a weighting coefficient for color difference. This refers to the degree of color difference.

[0075] The specific steps of the front-to-back image alignment operation are as follows: Deep learning was used to extract feature point sets from the highway images before and after maintenance, i.e., the feature point set before maintenance. and the set of characteristic points after maintenance In the formula: For the characteristic points before maintenance, These are characteristic points after maintenance. The number of feature points before maintenance. This represents the number of feature points after maintenance. Feature matching is performed based on the feature point sets before and after maintenance. The distance between feature descriptors is calculated, and the feature point with the smallest distance is selected as the matching point to obtain the correspondence between the highway images before and after maintenance.

[0076] The specific area restoration rate is as follows: By statistically analyzing the number of damaged pixels in highway images before and after maintenance, the damaged area before maintenance and the residual damaged area after maintenance are respectively used as the damaged area before maintenance and the residual damaged area after maintenance. The area repair rate is calculated based on the damaged area before maintenance and the residual damaged area after maintenance, using the following formula: ; In the formula: For area restoration rate, The area of ​​disease before maintenance. This refers to the area of ​​residual disease after maintenance.

[0077] Texture similarity is specifically calculated as follows: Texture features of the highway image before and after maintenance are obtained using the gray-level co-occurrence matrix and local binary mode, respectively. Texture similarity is then calculated based on these features, using the following formula: ; In the formula: For texture similarity, To preserve the texture features of the highway image before maintenance. Texture features of the highway image after maintenance.

[0078] Edge sharpness is specifically determined by obtaining the edge map of the highway image after maintenance using the Canny operator and calculating the ratio of the number of consecutive edge pixels to the total number of edge pixels.

[0079] The color difference is specifically calculated as follows: RGB data of the highway image after maintenance is obtained, converted to Lab data, and the color difference value between the maintained area and the remaining area is calculated using the following formula: ; In the formula: This represents the color difference value. For Lab data within the maintenance area, For the remaining Lab data within the region; The color difference degree is calculated based on the color difference value, using the following formula: ; In the formula: For color difference, This represents the color difference value. This is the preset maximum allowable color difference value.

[0080] In this embodiment of the invention, a highway pavement pothole repair project is used as the implementation object in a practical highway maintenance quality acceptance application scenario. After completing the discovery of the defects, dispatching maintenance orders, and on-site repair work, it is necessary to automatically compare and analyze the road condition before and after maintenance to determine whether the defect repair effect meets the acceptance requirements. Therefore, a highway image registration algorithm is used to spatially align the highway images before and after maintenance, and a maintenance quality evaluation model is constructed based on multiple dimensions such as area repair rate, texture similarity, edge clarity, and color difference to achieve automated and quantitative acceptance of the maintenance effect.

[0081] Acquire images of the highway before and after maintenance. The pre-maintenance images are collected by road inspection vehicles, drone inspection equipment, or fixed monitoring equipment during the defect discovery phase. These images contain complete status information of the defect area to be repaired, including the location, extent, and surface texture features of the defect. The post-maintenance images are re-acquired by inspection equipment after construction is completed, reflecting the road condition after repair. Based on the maintenance work order number, road marker number, and GPS positioning information, the pre-maintenance and post-maintenance images are correlated to form before-and-after image data pairs for the same area and the same defect location.

[0082] Because images before and after maintenance may differ due to variations in shooting angle, vehicle position shifts, lighting conditions, and image scale, preprocessing is performed on the acquired highway images. This preprocessing includes image size normalization, image enhancement, and defect region extraction. Image size normalization adjusts the two images to the same size, ensuring a consistent pixel space structure and reducing matching errors caused by size differences. Image enhancement improves the visibility of road surface texture and edge features by reducing noise and enhancing image contrast, thereby improving the accuracy of subsequent feature extraction. Defect region extraction utilizes the defect detection results generated before maintenance to mark defect areas in the image, obtaining defect region masks to provide a data basis for subsequent area repair rate calculations.

[0083] After image preprocessing, the alignment of the highway images before and after maintenance is further performed. Specifically, a deep learning feature extraction network is used to perform feature analysis on the highway images before and after maintenance, extracting key feature points from the two images, and constructing feature point sets before and after maintenance respectively.

[0084] After obtaining the feature point sets of the pre- and post-maintenance images, the feature descriptor distances between different feature points are calculated, and the corresponding matching relationships are determined by comparing the similarity of the feature descriptors. Specifically, for each feature point in the pre-maintenance image, the matching point with the smallest distance is found in the feature point set of the post-maintenance image. The smaller the distance between two feature points, the stronger their correspondence. After obtaining multiple valid matching points through screening, a spatial mapping relationship is established between the pre-maintenance and post-maintenance highway images, and image spatial transformation is performed based on the matching results to ensure that the road area, damaged area, and repaired area in the pre- and post-maintenance images remain spatially consistent.

[0085] After image alignment is completed, the highway images before and after maintenance are captured and evaluated for repair indicators, including area restoration rate, texture similarity, edge sharpness, and color difference.

[0086] Calculate the area restoration rate. Count the number of pixels in the damaged areas in both the pre-maintenance and post-maintenance highway images. The number of pixels in the damaged areas before maintenance is taken as the pre-maintenance damaged area, and the number of pixels in the damaged areas remaining after maintenance is taken as the post-maintenance residual damaged area. The smaller the post-maintenance residual damaged area, the higher the area restoration rate, indicating a more significant repair effect from the construction unit.

[0087] Texture similarity was calculated. Since road repair requires not only eliminating damaged areas but also ensuring high consistency between the repaired area and the original road surface, Gray-Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP) were used to extract texture features from the images before and after repair, respectively. This yielded texture feature vectors before and after repair, and cosine similarity was used to calculate the degree of texture similarity. The closer the texture structure of the repaired area is to the original road texture, the higher the calculated texture similarity, indicating a better integration effect between the repaired area and the surrounding road surface.

[0088] Edge sharpness is calculated. The Canny edge detection operator is used to process the post-maintenance highway image to extract edge information of the repaired area, and the ratio between the number of continuous edge pixels and the total number of edge pixels is calculated. When the edges of the repaired area are continuous, smooth, and without obvious cracks or abrupt changes, the edge sharpness score is improved.

[0089] Calculate color difference. Acquire RGB data of the highway image after maintenance and convert it to the Lab color space. Calculate the color distance by comparing the color difference between the repaired area and the surrounding normal road area. Further calculate the color evaluation index based on the color difference value. The closer the color of the repaired area is to the color of the surrounding road, the smaller the color difference value, and the higher the corresponding color evaluation value.

[0090] After obtaining the aforementioned multiple repair assessment indicators, each indicator is input into a multi-dimensional scoring algorithm to calculate the final maintenance quality score. For example, in practical applications, a scoring range is set according to the road maintenance acceptance standards. When the calculated maintenance quality score is greater than the upper limit of the preset scoring range, the current maintenance task is determined to have reached the first-level acceptance level, automatically passes the acceptance, and the acceptance result is written into the maintenance database. When the maintenance quality score is within the preset scoring range, the current task is determined to be at the second-level acceptance level, and a manual review task is generated for management personnel to further confirm the repair effect. When the maintenance quality score is lower than the lower limit of the preset scoring range, the current task is determined to be at the third-level acceptance level, automatically triggering the rework process and regenerating the maintenance task.

[0091] This invention utilizes highway image registration technology to achieve precise matching of road conditions before and after maintenance. It also combines multi-dimensional indicators such as area, texture, edge, and color to comprehensively evaluate maintenance quality. Compared to traditional acceptance methods relying on manual photo review or on-site inspection, this approach reduces human error, improves acceptance efficiency, and automates, standardizes, and automates highway maintenance quality evaluation. Furthermore, the acceptance results can be fed back to the maintenance management system, providing data support for construction unit evaluation, maintenance strategy optimization, and disease lifecycle management.

[0092] Example 2: A closed-loop management system for the entire lifecycle of highway maintenance, comprising a multi-source sensing module, a data confidence analysis module, an intelligent dispatching module, a construction tracking module, a maintenance quality assessment module, a strategy optimization engine, and a map management module; the multi-source sensing module is connected to the data confidence analysis module, the data confidence analysis module is connected to the intelligent dispatching module, the intelligent dispatching module is connected to the construction tracking module, the construction tracking module is connected to the maintenance quality assessment module, the maintenance quality assessment module is connected to the strategy optimization engine, and the strategy optimization engine is connected to the map management module.

[0093] The multi-source sensing module collects multi-source highway disease data through multi-source sensing devices, and uses image recognition algorithms to automatically identify the type, location and severity of highway diseases, and generates standardized disease records to the disease database. The data confidence analysis module is used to construct a road defect confidence analysis mechanism based on multi-source road defect data and to evaluate the accuracy of road defect identification. The intelligent work order dispatching module is used to dynamically generate the optimal maintenance work order dispatching strategy based on reinforcement learning algorithms, taking into account the severity of the disease and maintenance constraints, and to generate and push electronic work orders to the maintenance construction unit. The construction tracking module is used to track the maintenance construction process in real time, record attendance, progress reports and construction data, and support abnormal warnings and dynamic secondary dispatching. The maintenance quality assessment module is used to align the highway image capture and repair assessment indicators before and after maintenance using a highway image registration algorithm, and to calculate the maintenance quality score and determine the acceptance level through a multi-dimensional scoring algorithm. The strategy optimization engine uses maintenance quality scores as feedback signals to drive the update of reinforcement learning model parameters and automatically optimize dispatch rules and handling strategies. The map management module is used to construct a disease life cycle map, record the evolution trajectory of diseases at the same location and the effects of previous treatments, and make preventive maintenance decisions.

[0094] like Figure 2 The diagram shown is a system block diagram of a closed-loop management system for the entire life cycle of highway maintenance according to an embodiment of the present invention, which can be used to execute... Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0095] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of a closed-loop management method for the entire life cycle of highway maintenance as described in any of the preceding claims.

[0096] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can reside in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0097] This application also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the device to implement the methods provided in the various embodiments described above.

[0098] In the embodiments of the above-described device, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0099] Through the above embodiments, this invention achieves end-to-end management from disease discovery, dispatching and handling, construction tracking to acceptance evaluation, solving the problem of broken management chains in traditional systems and ensuring that each disease is traceable and assessable. Compared with traditional maintenance methods that rely on manual inspection and fixed-rule dispatching, this method uses multi-source data fusion technology to improve the accuracy and timeliness of disease identification, reduces misidentification and missed identification in complex environments through a confidence evaluation mechanism, and objectively and quantitatively evaluates the repair effect through image registration and quality scoring algorithms, solving the problems of strong subjectivity and inconsistent standards in manual acceptance, and improving acceptance quality and credibility. Based on reinforcement learning algorithms, the dispatching strategy is dynamically optimized by comprehensively considering multiple factors, significantly improving dispatching efficiency and resource utilization compared to fixed rules, and reducing maintenance costs. By employing AI image registration and multi-dimensional quality scoring algorithms, the system achieves objective and quantitative evaluation of maintenance effectiveness, reducing subjective errors caused by manual acceptance and improving the consistency of acceptance standards. Through a strategy self-optimization mechanism, the system continuously learns from historical maintenance data, automatically adjusting dispatch rules and handling strategies to continuously optimize management effectiveness over time. Real-time monitoring of the construction process and anomaly early warning mechanisms enable timely detection and dynamic adjustment of task execution deviations. Analysis of historical disease evolution patterns based on disease lifecycle mapping enables a shift from passive maintenance to proactive predictive maintenance, reducing repetitive maintenance costs and improving highway operational safety and the level of intelligent maintenance management. By accumulating a large amount of maintenance data and analysis results, the system provides data support for management decisions, reducing reliance on experience and improving the scientific rigor and accuracy of decision-making.

[0100] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

[0101] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A closed-loop management method for the entire life cycle of highway maintenance, characterized in that, Includes the following steps: Multi-source highway damage data are collected through multi-source sensing devices, and image recognition algorithms are used to automatically identify the type, location and severity of highway damage, generating standardized damage records for the damage database. A road defect confidence analysis mechanism was constructed based on multi-source road defect data to evaluate the accuracy of road defect identification. Based on reinforcement learning algorithms, the optimal maintenance dispatch strategy is dynamically generated by considering the severity of the disease and maintenance constraints, and electronic work orders are generated and pushed to the maintenance construction unit. Real-time tracking of the maintenance and construction process, recording on-site attendance, progress reports and construction data, and supporting abnormal warnings and dynamic secondary dispatching; A highway image registration algorithm is used to align the highway image capture and repair evaluation indicators before and after maintenance. A multi-dimensional scoring algorithm is used to calculate the maintenance quality score and determine the acceptance level. The maintenance quality score is used as a feedback signal to drive the update of reinforcement learning model parameters and automatically optimize dispatch rules and handling strategies. By constructing a disease life cycle map, the evolution trajectory of diseases at the same location and the effects of previous treatments are recorded, enabling preventive maintenance decisions.

2. The closed-loop management method for the entire life cycle of highway maintenance according to claim 1, characterized in that, A road defect confidence analysis mechanism was constructed based on multi-source road defect data to evaluate the accuracy of road defect identification. The specific steps are as follows: A multi-source highway defect data was acquired to establish a defect confidence analysis mechanism and calculate confidence analysis indicators. A weighted model is constructed based on confidence analysis indicators to obtain the comprehensive road damage confidence level. The comprehensive road damage confidence level is then compared with a preset confidence threshold to determine whether the confidence level of the corresponding road damage data is qualified. If the comprehensive road damage confidence level is greater than or equal to the preset confidence threshold, the confidence level of the corresponding road damage data is qualified. If the overall confidence level of the road defects is less than the preset confidence threshold, then the confidence level of the corresponding road defect data is unqualified. The accuracy of highway defect identification is evaluated based on the confidence level of highway defect data. The accuracy of highway defect identification is the ratio of the number of highway defect data with qualified confidence to the total number of highway defect data.

3. The closed-loop management method for the entire life cycle of highway maintenance according to claim 2, characterized in that, The confidence analysis metrics include image recognition confidence, spatial consistency confidence, and temporal continuity confidence.

4. The closed-loop management method for the entire life cycle of highway maintenance according to claim 1, characterized in that, Based on reinforcement learning algorithms, and considering both the severity of the damage and maintenance constraints, the optimal maintenance dispatch strategy is dynamically generated, and electronic work orders are generated and pushed to the maintenance construction unit. The specific steps are as follows: Obtain the severity of the disease and the maintenance constraints; the maintenance constraints include the conditions of the maintenance construction unit and the environmental conditions. Based on reinforcement learning algorithms, maintenance task actions are constructed according to the current highway status. By setting up a multi-dimensional maintenance task assignment reward function, the merits and demerits of maintenance task actions can be evaluated from multiple dimensions.

5. A closed-loop management method for the entire life cycle of highway maintenance according to claim 4, characterized in that, The constraints on the maintenance and construction unit's status include the current number of maintenance and construction tasks, the level of maintenance and construction technical capabilities, historical acceptance scores, and distance from the location of the defects; the constraints on the environment include weather conditions, road traffic flow, and the available time window for construction.

6. The closed-loop management method for the entire life cycle of highway maintenance according to claim 5, characterized in that, The multi-dimensional maintenance dispatch reward function includes rewards for timely treatment of defects, construction quality, cost optimization, and traffic impact.

7. The closed-loop management method for the entire life cycle of highway maintenance according to claim 1, characterized in that, A highway image registration algorithm was used to align the highway images captured before and after maintenance to assess the restoration indicators. A multi-dimensional scoring algorithm was then used to calculate the maintenance quality score and determine the acceptance level. The specific steps are as follows: Before and after highway maintenance images were acquired, and image preprocessing and alignment operations were performed, including image size normalization, image enhancement, and extraction of diseased areas. Repair assessment indicators were captured from aligned pre-maintenance and post-maintenance highway images. The maintenance quality score is calculated based on the repair assessment indicators input into a multi-dimensional scoring algorithm to determine the acceptance level. The maintenance quality score is compared with a preset scoring range to determine the acceptance level. When the maintenance quality score is greater than the upper limit of the preset scoring range, it is at the first-level acceptance level and automatically passes the acceptance. When the maintenance quality score is within the preset scoring range, it is at the second-level acceptance level and requires manual review. When the maintenance quality score is less than the lower limit of the preset scoring range, it is at the third-level acceptance level and triggers the rework process.

8. A closed-loop management method for the entire life cycle of highway maintenance according to claim 7, characterized in that, The restoration evaluation indicators include area restoration rate, texture similarity, edge sharpness, and color difference.

9. A closed-loop management system for the entire life cycle of highway maintenance, applied to the closed-loop management method for the entire life cycle of highway maintenance as described in any one of claims 1-8, characterized in that, It includes a multi-source sensing module, a data confidence analysis module, an intelligent dispatching module, a construction tracking module, a maintenance quality assessment module, a strategy optimization engine, and a map management module; The multi-source sensing module collects multi-source highway disease data through multi-source sensing devices, and uses image recognition algorithms to automatically identify the type, location and severity of highway diseases, and generates standardized disease records to the disease database. The data confidence analysis module is used to construct a road defect confidence analysis mechanism based on multi-source road defect data and to evaluate the accuracy of road defect identification. The intelligent work order dispatching module is used to dynamically generate the optimal maintenance work order dispatching strategy based on reinforcement learning algorithms, taking into account the severity of the disease and maintenance constraints, and to generate and push electronic work orders to the maintenance construction unit. The construction tracking module is used to track the maintenance construction process in real time, record attendance, progress reports and construction data, and support abnormal warnings and dynamic secondary dispatching. The maintenance quality assessment module is used to align the highway image capture and repair assessment indicators before and after maintenance using a highway image registration algorithm, and to calculate the maintenance quality score and determine the acceptance level through a multi-dimensional scoring algorithm. The strategy optimization engine uses maintenance quality scores as feedback signals to drive the update of reinforcement learning model parameters and automatically optimize dispatch rules and handling strategies. The map management module is used to construct a disease life cycle map, record the evolution trajectory of diseases at the same location and the effects of previous treatments, and make preventive maintenance decisions.

10. A readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it is used to implement the steps of a closed-loop management method for the entire life cycle of highway maintenance as described in any one of claims 1-8.

Images