A communication engineering construction site anomaly detection method and system based on visual recognition

Abstract

The application relates to the field of visual identification, and discloses a communication engineering construction site abnormality detection method and system based on visual identification, wherein the communication engineering construction site abnormality detection system based on visual identification comprises a video monitoring system and a monitoring and dispatching system; the monitoring and dispatching system comprises a monitoring and dispatching information acquisition unit used for acquiring the associated information of monitoring and dispatching tasks and a monitoring and dispatching unit used for generating a monitoring and dispatching plan; the monitoring and dispatching unit comprises a monitoring task identification module and a dispatching plan generation module; the task identification module is used for identifying the monitoring task of the next dispatching period from the associated information of the monitoring and dispatching tasks; the application can intelligently allocate dynamic monitoring tasks according to the real-time changes of the construction site and various environmental data, so as to ensure that key links are preferentially monitored.

Description

Technical Field

[0001] This invention relates to the field of visual recognition technology, and more specifically, to a method and system for detecting anomalies at communication engineering construction sites based on visual recognition. Background Technology

[0002] On-site monitoring and management of telecommunications engineering projects are essential to ensure construction safety, quality, and timely project completion. Traditional monitoring methods typically rely on manual inspections and fixed camera systems, which have the following shortcomings:

[0003] Insufficient real-time performance: Fixed monitoring equipment cannot fully cover dynamic and rapidly changing construction areas, resulting in delayed monitoring information.

[0004] Waste of resources: Manual inspection requires a lot of manpower and time, which is inefficient and unsustainable.

[0005] Lack of intelligence: Traditional monitoring systems usually rely on preset plans and lack the ability to adjust according to the real-time situation on site. Summary of the Invention

[0006] This invention provides a method and system for detecting anomalies at communication engineering construction sites based on visual recognition, solving the technical problem in related technologies where monitoring relies on preset schemes and cannot be flexibly adjusted.

[0007] This invention provides a vision-based anomaly detection system for communication engineering construction sites, including a video monitoring system. The video monitoring system includes an image acquisition unit installed at the monitoring point, an image storage device for storing the images acquired by the image acquisition unit, and further includes:

[0008] A monitoring and scheduling system includes a monitoring and scheduling information acquisition unit for acquiring associated information of monitoring and scheduling tasks and a monitoring and scheduling unit for generating monitoring and scheduling plans. The monitoring and scheduling unit includes a monitoring task identification module and a scheduling plan generation module. The task identification module is used to identify the monitoring tasks for the next scheduling cycle from the associated information of the monitoring and scheduling tasks. The scheduling plan generation module is used to analyze the associated information of the monitoring and scheduling tasks and the monitoring tasks for the next scheduling cycle to obtain the monitoring and scheduling plan for the next scheduling cycle. The monitoring and scheduling plan includes the location of the monitoring points for the next scheduling cycle and / or the monitoring tasks to be performed at each monitoring point.

[0009] Furthermore, the associated information for monitoring and scheduling tasks includes the task plan for communication engineering, information on monitoring tasks implemented in historical scheduling cycles, information on monitoring points in historical scheduling cycles, and information on sub-tasks of communication engineering carried out in historical scheduling cycles.

[0010] Furthermore, the task plan of a communications engineering project is an ordered arrangement of all sub-tasks of the communications engineering project, and information on the order in which all sub-tasks are implemented.

[0011] Furthermore, the associated information of the monitoring and scheduling tasks is processed in a structured manner before being input into the monitoring and scheduling unit to obtain ordered information in two structural dimensions. The basic structural elements contained in the first structural dimension are objects, including: monitoring points, communication links, monitoring tasks, sub-tasks, and construction areas; the basic structural elements contained in the second structural dimension are time-series units.

[0012] Furthermore, methods for processing ordered information include:

[0013] Objects from which ordered information is extracted;

[0014] Monitoring point object: Represents a monitoring point;

[0015] Communication link object: Represents the communication link between monitoring points or the communication link between a monitoring point and an image storage device;

[0016] Monitoring task object: Represents a monitoring task;

[0017] Subtask object: Represents a subtask of a communication project;

[0018] Construction area object: Represents a construction area;

[0019] Extract specific features associated with the object and divide the extracted specific features into multiple parts according to the scheduling cycle. Each part is mapped to a time unit, and the order of the time units is consistent with the time order of the scheduling cycle of the specific features.

[0020] Establish a mapping between the extracted objects and their associated specific features;

[0021] For any two objects, if there is a correlation between the specific characteristics associated with them that relates to the monitoring and scheduling task, or if there is a correlation between the objects themselves that relates to the monitoring and scheduling task, then an object relationship is established between the two objects.

[0022] Furthermore, the correlations related to monitoring and scheduling tasks include:

[0023] The monitoring points are adjacent in spatial location;

[0024] The monitoring point is connected to the communication link;

[0025] The image acquisition unit at the monitoring point performs monitoring tasks;

[0026] The relationship between monitoring tasks and subtasks;

[0027] The monitoring point is located within the construction area;

[0028] The relationship between sub-tasks and construction areas.

[0029] Furthermore, a deep learning model is used as the monitoring and scheduling unit. This deep learning model includes a first graph network layer, a first temporal layer, a first output layer, a second graph network layer, a second temporal layer, a feature fusion layer, and a second output layer. The first graph network layer takes into account the association information of the monitoring and scheduling tasks and outputs a first hidden feature to the first temporal layer. The first temporal layer outputs a second hidden feature to the first output layer and the feature fusion layer. The second graph network layer takes into account the association information of the monitoring and scheduling tasks and outputs a third hidden feature to the second temporal layer. The second temporal layer outputs a fourth hidden feature to the feature fusion layer. The feature fusion layer outputs the fused feature to the second output layer. The first output layer outputs the monitoring task results for the next scheduling cycle, and the second output layer outputs the monitoring and scheduling plan results for the next scheduling cycle.

[0030] Furthermore, the first graph network layer, the first temporal layer, and the first output layer constitute the task recognition module;

[0031] The second diagram consists of a network layer, a second temporal layer, a feature fusion layer, and a second output layer, forming a scheduling plan generation module.

[0032] This invention provides a method for detecting anomalies at communication engineering construction sites based on visual recognition, comprising the following steps:

[0033] Obtain the associated information of the monitoring and scheduling tasks;

[0034] Identify the monitoring tasks for the next scheduling cycle from the associated information of the monitoring and scheduling tasks;

[0035] The monitoring and scheduling plan for the next scheduling cycle is obtained by analyzing the associated information of the monitoring and scheduling tasks and the monitoring tasks for the next scheduling cycle.

[0036] The monitoring schedule plan includes the location of monitoring points for the next scheduling cycle and / or the monitoring tasks to be performed at each monitoring point;

[0037] Execute the monitoring and scheduling plan and deploy image acquisition units;

[0038] It reads image data acquired by the image acquisition unit and identifies the image data to perform monitoring tasks.

[0039] The present invention provides a computer-readable storage medium for storing computer-readable instructions, which, when read by a computer, can run the aforementioned visual recognition-based communication engineering construction site anomaly detection system.

[0040] The beneficial effects of the present invention are as follows: the present invention can intelligently allocate dynamic monitoring tasks based on real-time changes at the construction site and various environmental data, so as to ensure that key links are monitored first.

[0041] In a changing environment and with multiple possible locations, selecting the optimal deployment location for monitoring points is crucial for better executing future monitoring tasks and improving monitoring effectiveness. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of a visual recognition-based anomaly detection system for communication engineering construction sites according to the present invention.

[0043] Figure 2 This is an example of a task plan for a communications engineering project;

[0044] Figure 3 These are examples of monitoring tasks implemented during historical scheduling cycles;

[0045] Figure 4 These are examples of monitoring points in historical scheduling cycles;

[0046] Figure 5 This is an example of the division of monitoring areas;

[0047] Figure 6 This is an example of a matrix output by the second output layer;

[0048] Figure 7 This is another example of a matrix output by the second output layer. Detailed Implementation

[0049] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0050] First, it should be noted that using image information to identify image content and perform monitoring tasks is a common technique. It includes a video surveillance system 101 for acquiring images and an algorithm for identifying image content. The algorithm for identifying image content includes mature algorithms or models such as YOLO (You Only Look Once).

[0051] The application of video surveillance system 101 in communication engineering construction includes multiple aspects such as progress, quality, safety, cost, and collaboration;

[0052] Among the security applications are:

[0053] Hazard analysis at the construction site:

[0054] Construction sites for telecommunications projects present various hazards, such as working at heights, electrical work, and mechanical equipment operation. Analyzing these hazards helps workers better understand the safety risks at the construction site, enabling them to take appropriate preventative measures.

[0055] Safety of working at heights:

[0056] Working at heights is a common activity in telecommunications construction projects, and it is also a significant hazard on construction sites. When performing such work, workers must strictly adhere to relevant safety operating procedures and wear appropriate protective equipment such as safety belts and helmets to ensure their own safety.

[0057] Electrical work safety:

[0058] Electrical work involves electrical equipment and power lines, posing a risk of electric shock. When performing electrical work, workers must ensure that equipment is de-energized and wear protective equipment such as insulated gloves to prevent electric shock accidents.

[0059] Safety of mechanical equipment operation:

[0060] Commonly used machinery in telecommunications construction projects includes cranes and suspended platforms. Improper operation of these devices can easily cause accidental injuries. When operating such machinery, workers must receive professional training, be familiar with operating procedures, and ensure the safe and stable operation of the equipment.

[0061] The video surveillance system 101 can capture the dynamics of the construction site in real time, providing project managers and relevant managers with detailed information about the construction progress, helping them to better track and control the project.

[0062] The real-time footage recorded by the 101 video surveillance system can be used to track construction progress against the construction plan. Managers can review the surveillance video to confirm whether construction is proceeding according to schedule or to identify any deviations. To improve efficiency, intelligent video analytics technology can be employed. This technology uses algorithms to analyze video content, automatically identify discrepancies between construction progress and the plan, and generate corresponding reports.

[0063] The header of a list of examples for construction progress identification is as follows:

[0064] |Serial Number|Construction Area|Task Description|Start Time|End Time|Efficiency Assessment|.

[0065] Project quality is a core issue in construction management. Utilizing video surveillance technology allows for the timely detection of quality problems during construction, ensuring that construction quality meets standards.

[0066] Key points analysis of quality control:

[0067] Video surveillance can be used to monitor key quality control points during construction. For example, high-definition cameras can be installed for real-time monitoring of critical processes such as rebar tying and concrete pouring. Once a quality deviation is detected, construction personnel can be immediately notified for correction.

[0068] Traceability and handling of quality issues:

[0069] When quality issues arise, video surveillance data can be used as historical data for problem tracing. By analyzing the problematic stages, the causes can be identified, corresponding corrective measures can be taken, and preventative measures can be implemented for future construction. Utilizing video technology to review the construction process provides strong evidence for quality analysis and management.

[0070] Video surveillance can not only improve the visual management of construction quality and progress, but can also be directly applied to control project costs.

[0071] Construction of a cost monitoring indicator system:

[0072] Cost control requires establishing an indicator system that covers all key cost factors. For example, monitoring the usage of materials and equipment uptime. By combining video surveillance with sensor data, real-time monitoring of these key indicators can be achieved.

[0073] Resource allocation and scheduling optimization:

[0074] Video surveillance can provide information on resource utilization under actual construction conditions, including manpower, equipment, and materials. This can provide a basis for resource allocation and scheduling, reducing waste and optimizing costs.

[0075] At least one embodiment of the present invention discloses a visual recognition-based anomaly detection system for communication engineering construction sites, such as... Figure 1 As shown, it includes:

[0076] The video surveillance system 101 includes an image acquisition unit set at the monitoring point, an image storage device for storing the images acquired by the image acquisition unit, and a monitoring dynamic control unit for calculating and reasoning to obtain the optimal position of the monitoring point (the actual physical arrangement of the image acquisition unit still needs to be completed by manual or automated mechanical equipment).

[0077] The monitoring and scheduling system 102 includes a monitoring and scheduling information acquisition unit 1021 for acquiring associated information of monitoring and scheduling tasks and a monitoring and scheduling unit 1022 for generating a monitoring and scheduling plan. The monitoring and scheduling unit includes a monitoring task identification module 10221 and a scheduling plan generation module 10222. The task identification module 10221 is used to identify the monitoring tasks for the next scheduling cycle from the associated information of the monitoring and scheduling tasks. The scheduling plan generation module 10222 is used to analyze and obtain the monitoring and scheduling plan for the next scheduling cycle from the associated information of the monitoring and scheduling tasks and the monitoring tasks for the next scheduling cycle. The monitoring and scheduling plan includes at least the location of the monitoring points for the next scheduling cycle and may also include the monitoring tasks to be performed at each monitoring point.

[0078] In one embodiment of the present invention, the associated information of the monitoring and scheduling task includes the task plan of the communication project, information on the monitoring tasks implemented in the historical scheduling cycle, information on the monitoring points in the historical scheduling cycle, and information on the sub-tasks of the communication project carried out in the historical scheduling cycle.

[0079] The task plan for a communications engineering project can be an ordered arrangement of all subtasks, representing the order in which all subtasks are implemented. Alternatively, it can be ordered in the form of task clusters, where a task cluster contains multiple subtasks. When task clusters are ordered, the subtasks within the same task cluster are also ordered.

[0080] The information for a monitoring task includes the type of monitoring task, the description text of the monitoring task, and the monitoring results record of the monitoring task.

[0081] In one embodiment of the present invention, the monitoring task includes progress, quality, safety, cost, and type of collaboration;

[0082] It also includes hazard analysis at construction sites, safety of high-altitude operations, safety of electrical operations, safety of mechanical equipment operations, and types of resource usage.

[0083] Some of these types have a hierarchical relationship. For example, the safety type of monitoring tasks includes monitoring tasks for construction site hazard analysis, high-altitude operation safety, electrical operation safety, and mechanical equipment operation safety. However, it is understandable that a monitoring task may not have only one type.

[0084] In one embodiment of the present invention, the associated information of the monitoring and scheduling task is processed in a structured manner before being input into the monitoring and scheduling unit to obtain ordered information in two structural dimensions. The basic structural elements contained in the first structural dimension are objects, including: monitoring points, communication links, monitoring tasks, sub-tasks, and construction areas (which refer to the division of the area that can be affected by the construction of communication engineering, which can be a division based on sub-projects or a result of grid-based division); the basic structural elements contained in the second structural dimension are time-series units.

[0085] Methods for processing ordered information include:

[0086] Objects from which ordered information is extracted;

[0087] Monitoring point object: Represents a monitoring point;

[0088] Communication link object: Represents the communication link between monitoring points or the communication link between a monitoring point and an image storage device;

[0089] Monitoring task object: Represents a monitoring task;

[0090] Subtask object: Represents a subtask of a communication project;

[0091] Construction area object: Represents a construction area;

[0092] Extract specific features associated with the object and divide the extracted specific features into multiple parts according to the scheduling cycle. Each part is mapped to a time unit, and the order of the time units is consistent with the time order of the scheduling cycle of the specific features.

[0093] Establish a mapping between the extracted objects and their associated specific features;

[0094] Specific characteristics associated with monitoring points: the location of the monitoring point;

[0095] Specific characteristics associated with communication link objects:

[0096] The start and end points of a communication link: indicating the source and destination of the link connection.

[0097] Link bandwidth and latency: Performance metrics that describe the communication link.

[0098] Link status: Real-time reflection of link availability and health status.

[0099] Specific characteristics associated with the monitoring task object:

[0100] Monitoring task type: Specifies whether the monitoring task is of the progress, safety, quality, cost, or collaboration type.

[0101] Monitoring task priority: Priority determines the execution order or monitoring frequency of tasks.

[0102] Monitoring results: Results data or evaluation information obtained after task execution.

[0103] Task Description: Briefly describe the purpose and operational requirements of the monitoring task.

[0104] Specific characteristics associated with subtask objects:

[0105] Start and end times of subtasks: Identify the planned timeframe for each subtask.

[0106] Subtask dependencies: This describes the dependencies between the subtask and other tasks.

[0107] Resource requirements for subtasks: personnel, equipment, and materials required to execute subtasks.

[0108] Subtask completion status: Track the completion progress of subtasks in real time.

[0109] Specific characteristics associated with objects in the construction area:

[0110] Geographical location of the construction area: specific coordinates or location description of the construction area.

[0111] Construction environment conditions: Climate, geology and other environmental factors of the construction area.

[0112] Safety level of the construction area: safety risk assessment of the area.

[0113] Scope of construction impact: The surrounding environment and areas that may be affected by the construction in this area.

[0114] For any two objects, if there is an association between the specific characteristics of the associated objects that is related to the monitoring and scheduling task, or if there is an association between the objects that is related to the monitoring and scheduling task, then an object relationship is established between the two objects.

[0115] In some embodiments of the present invention, the correlations related to monitoring and scheduling tasks include:

[0116] The monitoring points are adjacent in spatial location;

[0117] The monitoring point is connected to the communication link;

[0118] The image acquisition unit at the monitoring point performs monitoring tasks;

[0119] The relationship between monitoring tasks and subtasks;

[0120] The monitoring point is located within the construction area;

[0121] The relationship between sub-tasks and construction areas.

[0122] In one embodiment of the present invention, the length of a scheduling cycle is 3-20 days, and the next monitoring scheduling plan is executed at the end of each scheduling cycle.

[0123] This scheduling cycle is generally proportional to the scale of the communication engineering construction, because the overall engineering environment changes more slowly in larger-scale construction, and the frequency of monitoring and scheduling can be lower.

[0124] In one embodiment of the present invention, a deep learning model is used as the monitoring and scheduling unit. The deep learning model includes a first graph network layer, a first temporal layer, a first output layer, a second graph network layer, a second temporal layer, a feature fusion layer, and a second output layer. The first graph network layer takes into account the association information of the monitoring and scheduling task and outputs a first hidden feature to the first temporal layer. The first temporal layer outputs a second hidden feature to the first output layer and the feature fusion layer. The second graph network layer takes into account the association information of the monitoring and scheduling task and outputs a third hidden feature to the second temporal layer. The second temporal layer outputs a fourth hidden feature to the feature fusion layer. The feature fusion layer outputs a fused feature to the second output layer. The first output layer outputs the result of the monitoring task for the next scheduling cycle, and the second output layer outputs the result of the monitoring and scheduling plan for the next scheduling cycle.

[0125] The task recognition module consists of the first network layer, the first temporal layer, and the first output layer.

[0126] The second diagram consists of a network layer, a second temporal layer, a feature fusion layer, and a second output layer, forming a scheduling plan generation module 10222;

[0127] In the foregoing embodiments of the present invention, although a monitoring and scheduling plan is provided, it does not exclude or prevent manual adjustment of the monitoring and scheduling plan.

[0128] In one embodiment of the present invention, both the first graph network layer and the second graph network layer adopt a multi-layer structure, and the calculation formula for the l-th layer is as follows:

[0129]

[0130] This represents the object recognition feature of the v-th object in the l-th layer. This represents the object recognition feature of the u-th object in the (l-1)-th layer. and Let each represent a set of objects that have object relationships with object v and object u, respectively. The cardinality of a set. This represents the planar recognition weight matrix of the l-th layer of the first data structure recognition layer. E represents the total number of floors, when l=1 , This represents the characteristic associated with the u-th object, when l=E. It equals the first or third hidden feature of object v (the first network layer outputs the first hidden feature, and the second network layer outputs the third hidden feature, but the trainable parameters of the first network layer and the second network layer are different). It is the sigmoid function.

[0131] Both the first and second temporal layers employ temporal neural networks, such as RNN, LSTM, or GRU.

[0132] One form of expression for the first and second time series layers is as follows:

[0133]

[0134]

[0135]

[0136]

[0137] and The activation vectors of the reset gate and update gate at step t;

[0138] : The candidate state generated in step t;

[0139] : The second or fourth hidden feature at step t (the first time series layer outputs the second hidden feature, and the second time series layer outputs the fourth hidden feature);

[0140] Transformation matrices 1, 2, 3, 4, 5, and 6 (trainable parameters);

[0141] : First, second, and third biases (trainable parameters);

[0142] r∈G, G represents the set of all objects, t∈{1,2,3,…,n}, when t=1 ;

[0143] , This represents the first hidden feature of object v output when the t-th temporal unit is input into the first graph network layer, or the third hidden feature of object v output when it is input into the second graph network layer. Represents a collection of all objects;

[0144] : Sigmoid activation function;

[0145] : tanh activation function;

[0146] The expression for the feature fusion layer is as follows:

[0147]

[0148] in Indicates fusion features, This represents the concatenation function. and These represent the second and fourth hidden features at step n, respectively, where n is the total number of temporal units.

[0149] In one embodiment of the present invention, the first output layer outputs a matrix, and a component of a row vector of the matrix represents the probability of a monitoring task in a task library being selected. The monitoring task with the highest probability is selected from the row vector as the monitoring task for the next scheduling cycle.

[0150] The task library refers to a library that stores monitoring tasks that may exist in the current communication project.

[0151] In one embodiment of the present invention, the construction area of ​​the entire communication project is discretized into multiple grids, and a monitoring point can be arranged in each grid. The location of the monitoring point can be represented by the ID of the grid.

[0152] The second output layer outputs a matrix. The c-th component of the i-th row vector of the matrix represents the probability of the i-th monitoring point selecting the c-th position. The position with the highest probability value is selected as the position of the i-th monitoring point in the next scheduling cycle.

[0153] Furthermore, the second output layer is divided into two layers. The first layer outputs a matrix, where the c-th component of the i-th row vector of the matrix represents the probability of the i-th monitoring point selecting the c-th position. The position with the highest probability value is selected as the position of the i-th monitoring point in the next scheduling cycle.

[0154] The second layer outputs a matrix. The c-th component of the i-th row vector of the matrix represents the probability that the i-th monitoring point selects the c-th monitoring task. The monitoring task with the highest probability value is selected as the monitoring task of the i-th monitoring point in the next scheduling cycle.

[0155] Data preparation:

[0156] When building such a deep learning model, the following types of data need to be prepared:

[0157] Geographic data: including gridded information of the construction area to facilitate the location of monitoring points.

[0158] Task data: A database of information about monitoring tasks, including task type, priority, etc.

[0159] Historical monitoring data: This includes monitoring tasks, their effects, and feedback within past scheduling cycles.

[0160] Task plan for communication engineering, such as Figure 2 As shown;

[0161] Examples of monitoring tasks implemented during historical scheduling cycles are as follows: Figure 3 As shown;

[0162] Examples of monitoring point information in historical scheduling cycles are as follows: Figure 4 As shown;

[0163] Examples of monitoring area division Figure 5 As shown.

[0164] Data preprocessing:

[0165] Data preprocessing is crucial for ensuring the accuracy and efficiency of deep learning models, including:

[0166] Data cleaning: Remove duplicate and erroneous data, and fill in missing values.

[0167] Normalization: Scaling numerical data to a common scale, such as [0,1], to improve the convergence of deep learning models.

[0168] Feature selection: Select features that have a greater impact on the prediction results based on the relevance and importance of the data.

[0169] Data splitting: Divide the dataset into training set, validation set and test set.

[0170] Deep learning model training:

[0171] The deep learning model first trains the task recognition module through supervised learning;

[0172] Then, the training schedule generation module 10222 is used through reinforcement learning. Reinforcement learning is suitable for dynamic decision-making problems and can learn and improve through interaction with the environment. The training process of a deep learning model is roughly as follows:

[0173] State definition: ordered information.

[0174] Action space: Available monitoring and scheduling plans.

[0175] Reward function: Design a reward mechanism based on optimizing the completion of monitoring tasks, maximizing efficiency, and minimizing resource usage to drive the deep learning model to learn in a direction that improves overall performance.

[0176] Reinforcement learning algorithm selection: such as Q-learning or Proximal Policy Optimization (PPO), etc.

[0177] Training process: The deep learning model iterates continuously in a simulation environment, gradually optimizing the strategy through trial and error.

[0178] like Figure 6 The image shows an example of a matrix output by the second output layer, such as... Figure 7 The image shows another example of the matrix output by the second output layer.

[0179] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A visual recognition-based anomaly detection system for communication engineering construction sites, comprising a video surveillance system, wherein the video surveillance system includes an image acquisition unit installed at a monitoring point and an image storage device for storing the images acquired by the image acquisition unit, characterized in that, Also includes: The monitoring and scheduling system includes a monitoring and scheduling information acquisition unit for acquiring the associated information of monitoring and scheduling tasks and a monitoring and scheduling unit for generating a monitoring and scheduling plan. The monitoring and scheduling unit includes a monitoring task identification module and a scheduling plan generation module. The task identification module is used to identify the monitoring tasks for the next scheduling cycle from the associated information of the monitoring and scheduling tasks. The scheduling plan generation module is used to analyze the associated information of the monitoring and scheduling tasks and the monitoring tasks for the next scheduling cycle to obtain the monitoring and scheduling plan for the next scheduling cycle. The monitoring schedule plan includes the location of monitoring points for the next scheduling cycle and / or the monitoring tasks to be performed at each monitoring point; The associated information for monitoring and scheduling tasks includes the task plan for communication engineering, information on monitoring tasks implemented in historical scheduling cycles, information on monitoring points in historical scheduling cycles, and information on sub-tasks of communication engineering carried out in historical scheduling cycles. The task plan for a communications engineering project is an ordered arrangement of all the sub-tasks of the project, and it contains information about the order in which all the sub-tasks are performed. The associated information of monitoring and scheduling tasks undergoes structured processing before being input into the monitoring and scheduling unit to obtain ordered information in two structural dimensions. The first structural dimension contains basic structural elements that are objects, including: monitoring points, communication links, monitoring tasks, subtasks, and construction areas. The second structural dimension contains basic structural elements that are time-series units. The methods for obtaining this ordered information include: Objects from which ordered information is extracted; Monitoring point object: Represents a monitoring point; Communication link object: Represents the communication link between monitoring points or the communication link between a monitoring point and an image storage device; Monitoring task object: Represents a monitoring task; Subtask object: Represents a subtask of a communication project; Construction area object: Represents a construction area; Extract specific features associated with the object and divide the extracted specific features into multiple parts according to the scheduling cycle. Each part is mapped to a time unit, and the order of the time units is consistent with the time order of the scheduling cycle of the specific features. Establish a mapping between the extracted objects and their associated specific features; For any two objects, if there is an association between the specific characteristics of the associated objects that is related to the monitoring and scheduling task, or if there is an association between the objects that is related to the monitoring and scheduling task, then an object relationship is established between the two objects. The correlations related to monitoring and scheduling tasks include: The monitoring points are adjacent in spatial location; The monitoring point is connected to the communication link; The image acquisition unit at the monitoring point performs monitoring tasks; The relationship between monitoring tasks and subtasks; The monitoring point is located within the construction area; The relationship between sub-tasks and construction areas; A deep learning model is used as the monitoring and scheduling unit. The deep learning model includes a first graph network layer, a first temporal layer, a first output layer, a second graph network layer, a second temporal layer, a feature fusion layer, and a second output layer. The first graph network layer takes into account the association information of the monitoring and scheduling tasks and outputs the first hidden feature to the first temporal layer. The first temporal layer outputs the second hidden feature to the first output layer and the feature fusion layer. The second graph network layer takes into account the association information of the monitoring and scheduling tasks and outputs the third hidden feature to the second temporal layer. The second temporal layer outputs the fourth hidden feature to the feature fusion layer. The feature fusion layer outputs the fused feature to the second output layer. The first output layer outputs the monitoring task results for the next scheduling cycle, and the second output layer outputs the monitoring and scheduling plan results for the next scheduling cycle. The task recognition module consists of the first network layer, the first temporal layer, and the first output layer. The second diagram consists of a network layer, a second temporal layer, a feature fusion layer, and a second output layer, which together form a scheduling plan generation module. Both the first and second graph network layers adopt a multi-layer structure. The calculation formula for the l-th layer is as follows: This represents the object recognition feature of the v-th object in the l-th layer. This represents the object recognition feature of the u-th object in the (l-1)-th layer. and Let each represent a set of objects that have object relationships with object v and object u, respectively. The cardinality of a set. This represents the planar recognition weight matrix of the l-th layer of the first data structure recognition layer. E represents the total number of floors, when l=1 , This represents the characteristic associated with the u-th object, when l=E. The first or third hidden feature of object v is equal to the first hidden feature. The first network layer outputs the first hidden feature, and the second network layer outputs the third hidden feature. However, the trainable parameters of the first and second network layers are different. It is the sigmoid function; Both the first and second temporal layers employ temporal neural networks; The expressions for the first and second time series layers are as follows: and The activation vectors of the reset gate and update gate at step t; : The candidate state generated in step t; The second or fourth hidden feature at step t: the second hidden feature is output by the first temporal layer, and the fourth hidden feature is output by the second temporal layer. Transformation matrices 1, 2, 3, 4, 5, and 6; : 1st, 2nd, and 3rd deviations; When t∈{1,2,3,…,n}, t=1 ; , This represents the first hidden feature of object v output when the t-th temporal unit is input into the first graph network layer, or the third hidden feature of object v output when it is input into the second graph network layer. Represents a collection of all objects; : Sigmoid activation function; : tanh activation function; The expression for the feature fusion layer is as follows: in Indicates fusion characteristics, This represents the concatenation function. and These represent the second and fourth hidden features at step n, respectively, where n is the total number of temporal units; The first output layer outputs a matrix. Each component of a row vector in the matrix represents the probability of a monitoring task in the task library being selected. The monitoring task with the highest probability is selected from the row vector as the monitoring task for the next scheduling cycle. The task library refers to a library that stores monitoring tasks that may exist in the current communication project.

2. A method for anomaly detection at a communication engineering construction site based on visual recognition, characterized in that, The following steps are performed based on the visual recognition-based communication engineering construction site anomaly detection system described in claim 1: Obtain the associated information of the monitoring and scheduling tasks; Identify the monitoring tasks for the next scheduling cycle from the associated information of the monitoring and scheduling tasks; The monitoring and scheduling plan for the next scheduling cycle is obtained by analyzing the associated information of the monitoring and scheduling tasks and the monitoring tasks for the next scheduling cycle. The monitoring schedule plan includes the location of monitoring points for the next scheduling cycle and / or the monitoring tasks to be performed at each monitoring point; Execute the monitoring and scheduling plan and deploy image acquisition units; It reads image data acquired by the image acquisition unit and identifies the image data to perform monitoring tasks.

3. A computer-readable storage medium, characterized in that, It is used to store computer-readable instructions, which, when read by a computer, enable the operation of the visual recognition-based communication engineering construction site anomaly detection system as described in claim 1.

Images