Well lid safety state artificial intelligence identification method and system for smart city construction

Abstract

The invention discloses a manhole cover safety state artificial intelligence identification method and system for smart city construction, and the method comprises the steps: obtaining manhole cover images and geographic position data at an edge calculation end, and carrying out the preliminary screening through a lightweight identification model; according to a screening result, carrying out shunting processing on the data, and uploading suspected abnormal data to a cloud server; and the cloud server carries out deep diagnosis by adopting a high-precision identification model and finally generates alarm information. The invention relates to the technical field of smart city infrastructure management. According to the method, the edge cloud collaborative architecture is constructed, high-speed preliminary screening is carried out at the edge end to filter mass normal data, and only a small amount of suspected abnormal data is uploaded to the cloud end to carry out fine diagnosis, so that the calculation load and network communication overhead of the edge end are greatly reduced, the contradiction between real-time performance and accuracy in manhole cover identification is effectively solved, and the manhole cover identification efficiency is improved. And the overall operation efficiency and economical efficiency of the system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart city infrastructure management, in particular to a manhole cover safety state artificial intelligence identification method and system for smart city construction. BACKGROUND

[0002] As a key node of the urban underground pipe network system, the manhole cover is an important part of the city infrastructure, and its safety state is directly related to road traffic safety and the safety of citizens. However, due to the large number of manhole covers, wide distribution, and long-term exposure to complex outdoor environments, they are prone to damage, displacement, loss, and other safety hazards. Traditional manhole cover management mainly relies on regular manual inspection, which has the problems of long inspection cycle, high labor cost, and delayed information feedback, making it difficult to achieve timely detection and disposal of sudden manhole cover hazards.

[0003] In recent years, with the development of artificial intelligence technology, the use of image recognition technology for automatic detection of manhole cover status has become a new research direction. These methods usually use deep learning models to analyze collected manhole cover images to identify abnormal states. However, existing technologies generally face a core challenge in large-scale applications, which is the contradiction between real-time identification and accuracy. On the one hand, to ensure high identification accuracy, complex and large parameter deep learning models are needed, which perform well on cloud servers, but if directly deployed on mobile inspection terminals (such as inspection vehicles, drones) with limited computing power, the inference speed is too slow to meet the real-time inspection requirements. On the other hand, if all high-definition video streams collected by the inspection terminal are uploaded to the cloud for real-time analysis, it will generate huge data transmission, causing great pressure on mobile communication networks (such as 4G / 5G), resulting in high communication costs and potential delays or instability affecting the timeliness of the alarm. This bottleneck of computing and communication resources severely restricts the deployment efficiency and economic feasibility of intelligent manhole cover monitoring systems.

[0004] Therefore, the present application provides a manhole cover safety state artificial intelligence identification method and system for smart city construction to solve the above problems. SUMMARY

[0005] To address the deficiencies of the prior art, the present application provides a manhole cover safety state artificial intelligence identification method and system for smart city construction, which solves the above problems.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: a manhole cover safety state artificial intelligence identification method for smart city construction, comprising:

[0007] At the edge computing end, obtain manhole cover image data and corresponding geographic location data;

[0008] At the edge computing end, a lightweight recognition model is used to preliminarily screen the manhole cover image data to generate a preliminary screening result.

[0009] According to the preliminary screening result, data shunting processing is performed on the manhole cover image data and corresponding geographic location data.

[0010] At the cloud server, a high-precision recognition model is used to perform deep diagnosis on the shunted data to generate an accurate diagnosis result.

[0011] Based on the accurate diagnosis result, manhole cover safety state alarm information is generated.

[0012] Preferably, the preliminary screening of the manhole cover image data using the lightweight recognition model comprises:

[0013] The manhole cover image data is identified as a preset "high probability normal" state or a "suspected abnormal" state.

[0014] Preferably, the data shunting processing according to the preliminary screening result comprises:

[0015] In response to the preliminary screening result being a "high probability normal" state, only the geographic location data and state log are uploaded to the cloud server;

[0016] In response to the preliminary screening result being a "suspected abnormal" state, the manhole cover image data and the geographic location data are uploaded to the cloud server.

[0017] Preferably, the data shunting processing is determined by an upload decision function, and the calculation formula is:

[0018]

[0019] Wherein: U is an upload decision variable, when U=1, the data upload operation of the "suspected abnormal" state is performed, when U=0, the data upload operation of the "high probability normal" state is performed; S edge is the confidence score output by the lightweight recognition model that the manhole cover image belongs to the "suspected abnormal" state, T edge is a preset confidence score threshold.

[0020] Preferably, the deep diagnosis using the high-precision recognition model comprises:

[0021] The manhole cover image data uploaded to the cloud server is accurately classified into at least one of the states of "damage", "missing", "uncovered" or "well circle problem".

[0022] Preferably, a work order containing problem type, geographical location and on-site image is automatically generated according to the severity level of the accurate diagnosis result, and is pushed to a maintenance management system.

[0023] Preferably, the method further comprises:

[0024] Obtaining manual verification feedback on the accurate diagnosis result;

[0025] Iteratively optimizing the lightweight recognition model and the high-precision recognition model using the manual verification feedback.

[0026] An intelligent well lid safety state artificial intelligence identification system for smart urban construction, comprising:

[0027] A data acquisition module for acquiring well lid image data and corresponding geographical location data at an edge computing end;

[0028] A preliminary screening module for preliminarily screening the well lid image data using a lightweight recognition model at the edge computing end to generate a preliminary screening result;

[0029] A data shunting module for shunting the well lid image data and corresponding geographical location data according to the preliminary screening result;

[0030] A deep diagnosis module for deep diagnosis of the shunted data using a high-precision recognition model at a cloud server to generate an accurate diagnosis result;

[0031] An alarm generation module for generating well lid safety state alarm information based on the accurate diagnosis result.

[0032] Preferably, the data shunting module comprises:

[0033] A log uploading unit for uploading only the geographical location data and state logs to the cloud server in response to the preliminary screening result being a "high probability of normal" state;

[0034] An abnormal data uploading unit for uploading the well lid image data and the geographical location data to the cloud server in response to the preliminary screening result being a "suspected abnormality" state.

[0035] Preferably, the system further comprises:

[0036] A visualization module for visualizing the safety state of the well lid on a geographic information system map according to the geographical location data and the accurate diagnosis result.

[0037] Advantages

[0038] The application provides a man-made intelligence identification method and system for the safety state of a manhole cover in smart city construction.

[0039] The application fundamentally solves the contradiction between real-time performance and accuracy by constructing a hybrid architecture of edge computing and cloud computing coordination, and by deploying a lightweight identification model on the edge computing end close to the data source to preliminarily screen a large number of manhole cover images at a high speed, and by quickly filtering out most manhole covers in a normal state. Only when a suspected abnormal manhole cover is found, the related data is selectively uploaded to the cloud. The cloud uses its powerful computing resources to deploy a high-precision identification model to finely diagnose a small amount of suspected abnormal data. This hierarchical processing mechanism greatly reduces the computing load of the edge end and the occupation of network communication bandwidth, ensures a high recall rate for potential hazards, realizes low-delay and high-precision diagnosis of abnormal events, and significantly improves the operation efficiency and economy of the whole system. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 A flowchart of the man-made intelligence identification method for the safety state of a manhole cover in smart city construction according to the embodiment of the application is shown.

[0041] Figure 2 A data flow diagram of the man-made intelligence identification method for the safety state of a manhole cover in smart city construction according to the embodiment of the application is shown.

[0042] Figure 3 A flowchart of data shunting processing according to the preliminary screening result according to the embodiment of the application is shown.

[0043] Figure 4 A block diagram of the man-made intelligence identification system for the safety state of a manhole cover in smart city construction according to the embodiment of the application is shown.

[0044] In the figure, 100 is a data acquisition module, 200 is a preliminary screening module, 300 is a data shunting module, 310 is a log uploading unit, 320 is an abnormal data uploading unit, 400 is a deep diagnosis module, 500 is an alarm generating module, and 600 is a visualization module. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0046] In the technical scheme of the present application, a man-made intelligence identification scheme for the safety state of a manhole cover based on edge computing and cloud computing is proposed to solve the technical problems of the prior art, such as the difficulty in balancing real-time performance and accuracy and the huge consumption of computing and communication resources. Specifically, in the present scheme, high-definition images and accurate geographic location information of the road manhole cover are synchronously obtained at the edge computing end such as a mobile inspection terminal. Subsequently, a lightweight identification model that has been compressed and optimized is used to perform high-speed reasoning on real-time video streams at the edge end, and the goal is not to accurately classify faults but to quickly identify the state of the manhole cover as being probably normal or suspected abnormal. The core is an intelligent data shunting mechanism: for the manhole cover that is judged to be normal, the system only records a brief log containing location information and uploads it in batches, thereby minimizing data transmission; for any manhole cover suspected to be abnormal, the system immediately uploads the corresponding high-definition images and metadata to the cloud server. After receiving the data, the cloud server calls a large-scale and high-precision identification model for deep diagnosis and outputs specific fault categories such as damage and loss. Finally, the system automatically generates an alarm work order based on the accurate diagnosis result of the cloud and pushes it to the city management platform, realizing an automatic closed loop from discovery, diagnosis to disposal.

[0047] In the technical scheme of the present application, a man-made intelligence identification scheme for the safety state of a manhole cover based on edge computing and cloud computing is proposed to solve the technical problems of the prior art, such as the difficulty in balancing real-time performance and accuracy and the huge consumption of computing and communication resources. Specifically, in the present scheme, high-definition images and accurate geographic location information of the road manhole cover are synchronously obtained at the edge computing end such as a mobile inspection terminal. Subsequently, a lightweight identification model that has been compressed and optimized is used to perform high-speed reasoning on real-time video streams at the edge end, and the goal is not to accurately classify faults but to quickly identify the state of the manhole cover as being probably normal or suspected abnormal. The core is an intelligent data shunting mechanism: for the manhole cover that is judged to be normal, the system only records a brief log containing location information and uploads it in batches, thereby minimizing data transmission; for any manhole cover suspected to be abnormal, the system immediately uploads the corresponding high-definition images and metadata to the cloud server. After receiving the data, the cloud server calls a large-scale and high-precision identification model for deep diagnosis and outputs specific fault categories such as damage and loss. Finally, the system automatically generates an alarm work order based on the accurate diagnosis result of the cloud and pushes it to the city management platform, realizing an automatic closed loop from discovery, diagnosis to disposal. Figure 1 A flowchart of the man-made intelligence identification method for the safety state of a manhole cover according to the embodiment of the present application is shown. Figure 2 A data flow diagram of the man-made intelligence identification method for the safety state of a manhole cover according to the embodiment of the present application is shown. Figure 1 And Figure 2 As shown in the drawings, the method comprises the steps of: S100, obtaining manhole cover image data and corresponding geographic location data at the edge computing end; S200, performing preliminary screening at the edge computing end using a lightweight identification model; S300, performing data shunting processing according to the preliminary screening result; S400, performing deep diagnosis at the cloud server using a high-precision identification model; and S500, generating alarm information based on the accurate diagnosis result.

[0048] Specifically, in step S100, the manhole cover image data and the corresponding geographic location data are obtained at the edge computing end. It should be understood that in order to effectively locate and dispose the manhole cover hazards, the visual appearance information and the spatial location information of the manhole cover must be obtained at the same time. The edge computing end usually refers to a computing device deployed on site, such as a city inspection vehicle, a drone or a smart monitoring pole.

[0049] More specifically, in one particular example of this application, a city patrol vehicle is equipped with an industrial-grade high-definition camera facing the road and a high-precision GPS module. The camera continuously acquires a 1080p resolution video stream of the road at a rate of 30 frames per second, forming manhole cover image data. Simultaneously, the GPS module synchronously records the vehicle's latitude and longitude coordinates at a frequency of 1Hz, forming geographic location data. An edge computing unit binds each frame of the image with the GPS coordinates closest to the timestamp, forming a complete data unit containing the image, timestamp, and geographic location, providing input for subsequent processing.

[0050] Specifically, in step S200, a lightweight recognition model is used at the edge computing end to perform preliminary screening of the manhole cover image data. It should be understood that the mobile inspection terminal has limited computing resources and cannot support real-time inference of a large-scale deep learning model. Therefore, in the technical solution of this application, a computationally efficient lightweight model is used as the front-end filter, whose core task is to quickly filter out massive amounts of normal manhole cover images while ensuring a high recall rate.

[0051] More specifically, in a concrete example of this application, the edge computing unit deploys a YOLOv5s model, converted to ONNX format and quantized with INT8, as a lightweight recognition model. This model performs frame-by-frame inference on the video stream acquired in the S100. The model's output is designed to have two categories: highly probable normal and suspected anomaly. The suspected anomaly category is a broad one, covering all potentially problematic manhole covers, including but not limited to low detection box confidence, incomplete manhole cover appearance, and obvious foreign objects or cracks in the manhole cover area. This design ensures that the system does not easily miss any potential risks.

[0052] Specifically, in step S300, data stream splitting is performed based on the preliminary screening results, which includes:

[0053] Step S310: In response to the initial screening result being "probably normal", only the geographic location data and status log are uploaded to the cloud server.

[0054] In step S320, in response to the preliminary screening result being in the "suspected abnormal" state, the manhole cover image data and geographical location data are uploaded to the cloud server. Figure 3 This is a detailed flowchart of the process. It should be understood that uploading all collected data indiscriminately would result in a huge waste of network bandwidth. Therefore, the technical solution of this application establishes an intelligent data routing mechanism to determine the data upload strategy based on the results of the initial screening, thereby achieving on-demand allocation of communication resources.

[0055] More specifically, in a particular example of this application, the data splitting logic is as follows:

[0056] If the output result of the lightweight model in S200 is a high probability of normal, the data shunting module only generates a simplified status log containing a timestamp, GPS coordinates, and a label indicating that the status is normal. These logs are cached locally and batched at a lower frequency (e.g., every 5 minutes) for uploading to the cloud database to form a complete inspection trajectory record.

[0057] If the output result is suspected abnormal, the data shunting module immediately packages the high-definition raw image of the frame, the corresponding geographic location data, the timestamp, and the preliminary screening label, and uploads them in real time to the designated data portal of the cloud server through the 5G network with high priority.

[0058] This shunting decision can be formally described by an upload decision function, whose calculation formula is as follows:

[0059]

[0060] Where: U is the upload decision variable, when U = 1, the data upload operation of the "suspected abnormal" state is performed, when U = 0, the data upload operation of the "high probability of normal" state is performed; S edge is the confidence score output by the lightweight recognition model that the manhole cover image belongs to the "suspected abnormal" state, T edge is the pre-set confidence score threshold.

[0061] Specifically, in step S400, the cloud server uses a high-precision recognition model to perform in-depth diagnosis on the data processed by the shunting. It should be understood that the preliminary screening result of the edge end is relatively rough, and a more powerful model is needed for accurate fault classification. The cloud server has strong computing power and is suitable for deploying complex high-precision models.

[0062] More specifically, in one specific example of the present application, the cloud server deploys a complete YOLOv8x model as a high-precision recognition model. This model receives the suspected abnormal data packet uploaded by the edge end. It performs inference on the high-definition image therein, outputs accurate fault categories such as damage, loss, uncapping, and well circle problems, and provides accurate bounding box coordinates for each fault area and a classification confidence of up to 95% or more.

[0063] Specifically, in step S500, based on the accurate diagnosis result, the manhole cover safety state alarm information is generated. It should be understood that the diagnosis result of the model needs to be converted into instructions for the operation and maintenance personnel to execute. Therefore, this step aims to automatically convert the technical diagnosis conclusion into a standardized business process event.

[0064] More specifically, in a concrete example of this application, when the S400's diagnostic results confirm one or more abnormal manhole cover conditions, the alarm generation module assesses their severity level according to preset rules. For example, missing and uncovered covers are defined as the highest priority. The system then automatically creates an alarm work order, which includes the fault type (e.g., missing manhole cover), precise GPS coordinates, severity level, discovery time, and high-resolution images of the site. This work order is pushed in real-time to the city's integrated management platform's work order system via API interface, or sent directly as a message to the mobile work terminal of the area maintenance personnel, thereby initiating the offline handling process.

[0065] Furthermore, this method may include a closed-loop step of model iterative optimization. After maintenance personnel complete the on-site handling, they feed back the verified actual fault type (e.g., confirmed well ring damage) to the system via mobile terminal. These manually verified images and their corresponding real labels will be automatically added to the training dataset as high-quality training samples. The system periodically uses this new data to retrain or fine-tune the high-precision model in the cloud and the lightweight model at the edge, thereby continuously improving the recognition performance of the entire system with ongoing use.

[0066] Furthermore, this application also provides an artificial intelligence recognition system for the safety status of manhole covers for smart city construction. Figure 4 This is a block diagram of an artificial intelligence recognition system for the safety status of manhole covers for smart city construction, according to an embodiment of this application. Figure 4 As shown, the system includes: a data acquisition module 100, used to acquire manhole cover image data and corresponding geographic location data at the edge computing end; a preliminary screening module 200, used to perform preliminary screening of the manhole cover image data at the edge computing end using a lightweight recognition model to generate preliminary screening results; a data diversion module 300, used to perform data diversion processing on the manhole cover image data and corresponding geographic location data based on the preliminary screening results. Internally, it can be further divided into a log upload unit and an abnormal data upload unit, handling normal and suspected abnormal situations respectively; a deep diagnosis module 400, used to perform deep diagnosis on the diverted data using a high-precision recognition model on the cloud server to generate accurate diagnostic results; and an alarm generation module 500, used to generate manhole cover safety status alarm information based on the accurate diagnostic results. The data diversion module 300 also includes: a log upload unit 310, which, in response to the initial screening result being "highly likely normal", uploads only the geographic location data and status log to the cloud server; and an abnormal data upload unit 320, which, in response to the initial screening result being "suspected abnormal", uploads the manhole cover image data and geographic location data to the cloud server.

[0067] In an optional embodiment, the system can further include a visualization module 600. The module is used to visualize all the manhole covers on a geographic information system (GIS) map according to their geographic location data and their latest status (whether normal reported by the edge side or abnormal diagnosed by the cloud side) with icons of different colors. This enables the management personnel to intuitively grasp the overall health status and risk distribution of the manhole covers in the city, providing data support for macro decision-making.

[0068] As described above, the manhole cover safety state artificial intelligence identification system for smart urban construction according to the embodiments of the present application can be implemented in various wireless terminals, such as a server with a manhole cover safety state artificial intelligence identification control algorithm for smart urban construction. In a possible implementation manner, the manhole cover safety state artificial intelligence identification system for smart urban construction according to the embodiments of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the manhole cover safety state artificial intelligence identification system for smart urban construction can be a software module in the operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the manhole cover safety state artificial intelligence identification system for smart urban construction can also be one of the many hardware modules of the wireless terminal.

[0069] Meanwhile, the contents not described in detail in the present specification all belong to the prior art known to those skilled in the art.

[0070] It should be noted that, in this paper, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0071] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A manhole cover safety state artificial intelligence identification method for smart urban construction, characterized in that, The method comprises: acquiring manhole cover image data and corresponding geographic location data at an edge computing end; preliminarily screening the manhole cover image data using a lightweight recognition model at the edge computing end to generate a preliminary screening result; performing data shunting processing on the manhole cover image data and the corresponding geographic location data according to the preliminary screening result; performing deep diagnosis on the data subjected to the shunting processing using a high-precision recognition model at a cloud server to generate an accurate diagnosis result; generating manhole cover safety state alarm information based on the accurate diagnosis result.

2. The manhole cover safety state artificial intelligence identification method for smart urban construction according to claim 1, characterized in that, The preliminary screening of the manhole cover image data using the lightweight recognition model comprises: identifying the manhole cover image data as a preset "high probability of normal" state or a "suspected abnormal" state.

3. The manhole cover safety state artificial intelligence identification method for smart urban construction according to claim 2, characterized in that, The data shunting processing according to the preliminary screening result comprises: in response to the preliminary screening result being the "high probability of normal" state, uploading only the geographic location data and state log to the cloud server; in response to the preliminary screening result being the "suspected abnormal" state, uploading the manhole cover image data and the geographic location data to the cloud server.

4. The manhole cover safety state artificial intelligence identification method for smart urban construction according to claim 3, characterized in that, The data shunting processing is determined by an uploading decision function, and the calculation formula is: wherein: U is an upload decision variable, when U = 1, a data upload operation in the "suspected abnormality" state is performed, when U = 0, a data upload operation in the "high probability of normality" state is performed; S edge is a confidence score output by the lightweight recognition model about whether the manhole cover image belongs to the "suspected abnormality" state, T edge is a preset confidence score threshold.

5. The manhole cover safety state artificial intelligence identification method for smart urban construction of claim 1, wherein, The deep diagnosis using the high-precision recognition model comprises: accurately classifying the manhole cover image data uploaded to the cloud server into at least one state of "damage", "loss", "uncovering" or "manhole ring problem".

6. The manhole cover safety state artificial intelligence identification method for smart urban construction of claim 1, wherein, According to the severity level of the accurate diagnosis result, a work order containing the problem type, geographic location and on-site image is automatically generated and pushed to a maintenance management system.

7. The manhole cover safety state artificial intelligence identification method for smart urban construction of claim 1, wherein, The method further comprises: acquiring manual verification feedback on the accurate diagnosis result; iteratively optimizing the lightweight recognition model and the high-precision recognition model using the manual verification feedback.

8. A manhole cover safety state artificial intelligence identification system for smart urban construction, characterized in that, The system comprises: a data acquisition module (100) configured to acquire manhole cover image data and corresponding geographic location data at an edge computing end; a preliminary screening module (200) configured to preliminarily screen the manhole cover image data using a lightweight recognition model at the edge computing end to generate a preliminary screening result; a data shunting module (300) configured to perform data shunting processing on the manhole cover image data and the corresponding geographic location data according to the preliminary screening result; a deep diagnosis module (400) configured to perform deep diagnosis on the data subjected to the shunting processing using a high-precision recognition model at a cloud server to generate an accurate diagnosis result; an alarm generation module (500) configured to generate manhole cover safety state alarm information based on the accurate diagnosis result.

9. The manhole cover safety state artificial intelligence identification system for smart urban construction according to claim 8, characterized in that, The data shunting module (300) comprises: a log uploading unit (310) configured to, in response to the preliminary screening result being the "high probability of normal" state, upload only the geographic location data and state log to the cloud server; an abnormal data uploading unit (320) configured to, in response to the preliminary screening result being the "suspected abnormal" state, upload the manhole cover image data and the geographic location data to the cloud server.

10. The manhole cover safety state artificial intelligence identification system for smart urban construction of claim 8, wherein, The system further comprises: A visualization module (600) is configured to visualize the safety status of the manhole cover on a geographic information system map according to the geographic position data and the accurate diagnosis result.

Images