Electric power infrastructure project monitoring method and system based on deep learning

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an electric power infrastructure project monitoring method and system based on deep learning, which are used for improving the efficiency and accuracy of electric power infrastructure project monitoring based on deep learning. The method comprises the following steps: performing model training on a Faster R-CNN algorithm through a labeled image set to obtain an electric power infrastructure project progress recognition model; collecting a monitoring video image of a preset target area, and inputting the monitoring video image into the electric power infrastructure project progress recognition model for progress recognition to obtain current stage data; performing construction scheme association mapping according to the current stage data to obtain a current project progress percentage, and predicting a project fund expenditure estimated value according to the current project progress percentage; and constructing a BIM digital twinborn model according to the current project progress percentage and the project fund expenditure estimated value, performing visual element construction on the BIM digital twinborn model to obtain a visual element set, and performing model display on a preset data display terminal through the visual element set.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a power infrastructure project monitoring method and system based on deep learning. Background Art

[0002] In the existing technology, the progress monitoring and management of power infrastructure projects usually rely on the traditional method of manual inspection and manual recording. Specifically, project managers need to dispatch special construction supervisors to visit scattered construction sites regularly, judge and record the actual progress of each construction area through visual observation, and manually input the collected progress data into the progress management system. This monitoring mode that relies entirely on manual work has many shortcomings.

[0003] The efficiency and accuracy of manual monitoring are low. The construction site environment is complex and the operation area is vast. It is difficult for limited supervisors to achieve all-weather and all-around monitoring of the entire operation area, which is prone to blind spots. Even experienced supervisors are inevitably affected by subjective factors and lead to deviations in progress judgment. The cost of manual monitoring is high. A large amount of human resources is required, which brings a heavy labor cost burden to the company. Moreover, the on-site working environment is poor and there are certain safety hazards, which brings potential risks to the personal safety of supervisors.

[0004] In addition, the traditional manual monitoring mode cannot be fully integrated with modern building information modeling (BIM) technology. BIM technology can build a digital twin model of the project and realize visual management of the entire life cycle. However, the existing technology lacks an effective method to combine construction progress data with the BIM model, and cannot use the BIM model to fully simulate the project and support decision-making, which brings certain blind spots to project management. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a method and system for monitoring electric power infrastructure projects based on deep learning, which are used to improve the efficiency and accuracy of monitoring electric power infrastructure projects based on deep learning.

[0006] The present invention provides a method for monitoring electric power infrastructure projects based on deep learning, comprising: collecting stage marker samples for multiple construction stages to obtain a marker sample set, wherein the marker sample set includes marker samples for each construction stage; inputting the marker sample set into a threshold segmentation algorithm for image annotation to obtain a labeled image set; performing model training on a preset Faster R-CNN algorithm through the labeled image set to obtain an electric power infrastructure project progress recognition model; real-time acquisition of monitoring video images of a preset target area, inputting the monitoring video images into the electric power infrastructure project progress recognition model for progress identification to obtain current stage data; performing construction scheme association mapping according to the current stage data to obtain the current project progress percentage, and predicting a project capital expenditure estimate according to the current project progress percentage; constructing a BIM digital twin model according to the current project progress percentage and the project capital expenditure estimate, constructing visualization elements for the BIM digital twin model to obtain a visualization element set, and displaying the model on a preset data display terminal through the visualization element set.

[0007] In the present invention, the step of collecting stage marker samples for multiple construction stages to obtain a marker sample set, wherein the marker sample set includes marker samples for each construction stage, includes: dividing multiple key areas in a preset historical construction site to obtain a key area set, wherein each key area in the key area set corresponds to a construction stage; collecting historical videos for each key area according to the key area set to obtain a historical monitoring video stream set; parsing the historical monitoring video stream set frame by frame to obtain a key frame image set, wherein the key frame image set includes key frame images of each key area; performing image enhancement processing on the key frame image set to obtain an enhanced image set; Perform image segmentation on the enhanced image set, extract the contours of construction objects in each image, and obtain a construction object contour set; perform template matching on the construction object contour set according to a preset construction stage marker template to obtain a matching result set; extract the construction object contours with the highest matching degree with the marker template according to the matching result set to obtain a matching contour set; perform size normalization on the matching contour set to obtain normalized marker samples; encode the normalized marker samples according to the sequence of construction stages to obtain a coded marker sample set; classify the coded marker sample set according to the construction stage to obtain the marker sample set, wherein the marker sample set includes marker samples for each construction stage.

[0008] In the present invention, the step of inputting the marker sample set into a threshold segmentation algorithm for image annotation to obtain an annotated image set includes: graying the marker sample set to obtain a grayscale marker sample set; performing histogram equalization on the grayscale marker sample set to obtain a grayscale marker sample set with enhanced contrast; binarizing the grayscale marker sample set with enhanced contrast according to a preset threshold to obtain a binary marker sample set; performing morphological opening operation on the binary marker sample set to obtain a binary marker sample set after the opening operation; performing connected domain analysis on the binary marker sample set after the opening operation to obtain A binary marker sample set with connected domain labels; filtering the binary marker sample set with connected domain labels according to a preset area threshold to obtain a filtered binary marker sample set; performing edge extraction on the filtered binary marker sample set to obtain a marker contour edge set; performing contour alignment on the original marker samples according to the marker contour edge set to obtain a marker sample set after contour alignment; performing size normalization on the marker sample set after contour alignment to obtain a size-normalized marker sample set; performing image annotation on the size-normalized marker sample set according to a preset annotation rule to obtain an annotated image set.

[0009] In the present invention, the real-time acquisition of monitoring video images of a preset target area, inputting the monitoring video images into the electric power infrastructure project progress recognition model for progress recognition, and obtaining the current stage data step include: numbering multiple monitoring cameras in the preset target area to obtain a camera number set, wherein each camera corresponds to a unique number; according to the camera number set, real-time acquisition of the video stream of each camera to obtain a multi-channel monitoring video stream set; synchronizing and aligning the multi-channel monitoring video stream set to obtain a synchronized monitoring video stream set; extracting target frames from the synchronized monitoring video stream set at preset time intervals to obtain a target frame image set, wherein each target frame image corresponds to a timestamp; performing image preprocessing on the target frame image set to obtain a preprocessed target frame image set. image set; input the preprocessed target frame image set into the electric power infrastructure project progress recognition model, perform feature extraction and target detection on each target frame image, and obtain a target frame detection result set; perform non-maximum suppression and bounding box merging processing on the target frame detection result set to obtain an optimized target frame detection result set; according to the optimized target frame detection result set, perform construction stage recognition on each target frame image to obtain a target frame construction stage recognition result set; sort the target frame construction stage recognition result set according to timestamps to obtain a time-series construction stage recognition result sequence; perform voting fusion on the time-series construction stage recognition result sequence, determine the construction stage corresponding to each timestamp according to the majority voting principle, and obtain current stage data, including the current construction stage number and the corresponding timestamp.

[0010] In the present invention, the step of performing image preprocessing on the target frame image set to obtain the preprocessed target frame image set includes: performing image denoising on the target frame image set by a median filtering algorithm to obtain a denoised target frame image set; performing image enhancement on the denoised target frame image set to obtain an enhanced target frame image set; performing edge-preserving filtering on the enhanced target frame image set by a bilateral filtering algorithm to obtain an edge-enhanced target frame image set; performing color space conversion on the edge-enhanced target frame image set to obtain a target frame image set in an HSV color space; performing channel data extraction on the target frame image set in the HSV color space to obtain a hue channel image set, a saturation channel image set, and a color space conversion on the target frame image set in an HSV color space. channel image set and a brightness channel image set; performing histogram statistics on the hue channel image set to obtain a hue histogram feature vector set, performing histogram statistics on the saturation channel image set to obtain a saturation histogram feature vector set, performing histogram statistics on the brightness channel image set to obtain a brightness histogram feature vector set; splicing the hue histogram feature vector set, the saturation histogram feature vector set and the brightness histogram feature vector set to obtain an HSV color histogram feature vector set; normalizing the HSV color histogram feature vector set to obtain a normalized HSV color histogram feature vector set; using the normalized HSV color histogram feature vector set as the preprocessed target frame image set.

[0011] In the present invention, the steps of performing construction scheme association mapping according to the current stage data to obtain the current engineering progress percentage and predicting the estimated value of project capital expenditure according to the current engineering progress percentage include: establishing a mapping relationship library between construction stages and construction tasks, wherein each construction stage corresponds to a plurality of specific construction tasks, and a construction stage-task mapping table is obtained; according to the current stage data, searching the construction stage-task mapping table for a list of completed construction tasks corresponding to the current construction stage to obtain a list of completed tasks; for each task in the completed task list, obtaining the corresponding planned engineering quantity from a pre-established construction task-engineering quantity mapping table, and counting the cumulative completed engineering quantities of all completed tasks to obtain the cumulative completed engineering quantity; and obtaining the cumulative completed engineering quantity from the construction task-engineering quantity mapping table. The total planned engineering quantity is obtained from the mapping table, and the cumulative completed engineering quantity is divided by the total planned engineering quantity to obtain the current engineering progress percentage; a mapping relationship library between construction tasks and cost expenditures is established to obtain a construction task-cost mapping table; cost expenditure analysis is performed based on the completed task list and the construction task-cost mapping table to obtain the cumulative actual cost expenditures of the completed tasks; budget cost expenditure analysis is performed on the completed task list based on the current engineering progress percentage to obtain the budget cost expenditures of the completed tasks; the difference between the cumulative actual cost expenditures of the completed tasks and the budget cost expenditures of the completed tasks is calculated to obtain the difference data; based on the difference data and the current engineering progress percentage, a forecast of the project capital expenditure is estimated to obtain the project capital expenditure estimate.

[0012] In the present invention, the BIM digital twin model is constructed according to the current engineering progress percentage and the estimated value of the project capital expenditure, visualization elements are constructed for the BIM digital twin model to obtain a visualization element set, and the model display step is performed on a preset data display terminal through the visualization element set, including: vector decomposition of the construction plan according to the current engineering progress percentage to obtain a project quantity feature vector set; feature mapping of the parameterized algorithm according to the project quantity feature vector set to obtain a resource consumption mapping set; cost accounting of the resource consumption mapping set to obtain cost budget data; and correction and adjustment of the estimated value of the project capital expenditure according to the cost budget data to obtain a corrected project The revised estimated value of project capital expenditure is input into the BIM three-dimensional model construction algorithm to obtain the original BIM digital twin model; the original BIM digital twin model is rendered with materials to obtain the BIM model after material rendering; the lighting simulation is performed according to the BIM model after material rendering to obtain the BIM model after lighting rendering; the viewing angle is set for the BIM model after lighting rendering to obtain the BIM visualization model after viewing angle setting; the BIM visualization model after viewing angle setting is animated to obtain the animated BIM visualization model; the animated BIM visualization model is output as a visualization element set for model display on a preset data display terminal.

[0013] The present invention also provides a power infrastructure project monitoring system based on deep learning, comprising: A collection module, used for collecting stage marker samples for multiple construction stages to obtain a marker sample set, wherein the marker sample set includes marker samples for each construction stage; A labeling module, used for inputting the marker sample set into a threshold segmentation algorithm for image labeling to obtain a labeling image set; A training module, used to perform model training on a preset Faster R-CNN algorithm using the annotated image set to obtain a power infrastructure project progress recognition model; An identification module is used to collect monitoring video images of a preset target area in real time, input the monitoring video images into the electric power infrastructure project progress identification model for progress identification, and obtain current stage data; A mapping module is used to perform construction scheme association mapping according to the current stage data, obtain the current project progress percentage, and predict the estimated value of project capital expenditure according to the current project progress percentage; A display module is used to build a BIM digital twin model according to the current engineering progress percentage and the estimated value of the project capital expenditure, construct visualization elements for the BIM digital twin model to obtain a visualization element set, and display the model on a preset data display terminal through the visualization element set.

[0014] In the technical solution provided by the present invention, by collecting marker samples for multiple construction stages, and inputting the samples into a threshold segmentation algorithm for image annotation, a set of annotated images is obtained, and then the Faster R-CNN algorithm is trained using the annotated image set to obtain a progress recognition model for power infrastructure projects. The real-time monitoring video images of the construction site can be accurately and efficiently identified for progress, and the current construction stage data can be automatically output. This progress monitoring method based on artificial intelligence visual recognition technology does not require manual on-site inspections and records, greatly improves monitoring efficiency, reduces human operational errors, and avoids safety hazards. According to the identified current stage data, the real-time progress percentage of the project is accurately calculated through association mapping with the construction plan, and the project capital expenditure is estimated based on the progress percentage, which provides a basis for project managers to timely understand the use of funds and adjust the capital allocation strategy, thereby achieving more refined and rationalized capital management and control. Combine the project progress data with BIM (Building Information Model) technology, build a BIM digital twin model of the project based on the progress percentage and funding estimate, and perform various forms of visualization on the model, including material rendering, lighting simulation, perspective setting, and animation, and finally output a full range of visual model element sets, which are intuitively displayed using a data display terminal. This BIM digital twin modeling technology enables project managers to clearly understand the implementation progress and funding consumption of the entire project, providing a scientific reference for subsequent decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 This is a flow chart of a method for monitoring electric power infrastructure projects based on deep learning in an embodiment of the present invention; Figure 2 Schematic diagram of a power infrastructure project monitoring system based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0019] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0020] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , Figure 1 is a flow chart of a method for monitoring electric power infrastructure projects based on deep learning according to an embodiment of the present invention, such as Figure 1 As shown, the following steps are included: S101, collecting stage marker samples for multiple construction stages to obtain a marker sample set; Among them, the marker sample set includes marker samples for each construction stage; S102, inputting the marker sample set into a threshold segmentation algorithm for image annotation to obtain an annotated image set; S103, training a preset Faster R-CNN algorithm through the labeled image set to obtain a power infrastructure project progress recognition model; wherein the Faster R-CNN algorithm is a second-order target detection algorithm based on deep learning; S104, collecting monitoring video images of a preset target area in real time, inputting the monitoring video images into a progress recognition model for power infrastructure projects to perform progress recognition, and obtaining current stage data; S105, performing construction scheme association mapping according to the current stage data, obtaining the current project progress percentage, and predicting the estimated value of project capital expenditure according to the current project progress percentage; S106. Build a BIM digital twin model based on the current engineering progress percentage and the estimated value of project capital expenditure, construct visualization elements for the BIM digital twin model to obtain a visualization element set, and display the model on a preset data display terminal through the visualization element set.

[0021] It should be noted that the historical construction site video data is analyzed and extracted. Specifically, the key areas of the construction site are first divided, each key area corresponds to a construction stage, and then the historical monitoring videos of these areas are collected, the video stream is parsed frame by frame, and the key frame image set is extracted. Next, the key frame images are processed using image enhancement, segmentation and other algorithms to extract the contours of the main construction objects in each image, and then the extracted contours are template matched according to the pre-set construction stage marker template to find the contour set that best matches the marker. Finally, the matching contours are normalized and coded and classified according to the construction order to obtain the marker sample set corresponding to each construction stage. Taking the construction of the power tower foundation as an example, the markers at different stages may be steel cages, concrete pouring, wire positioning piles, etc. Through the above processing, the typical contours of these objects can be automatically extracted from the historical video as marker samples for each stage. Next, the obtained marker sample set is input into the threshold segmentation algorithm for image annotation. Specifically, the sample image is first preprocessed by grayscale conversion and histogram equalization to improve the image contrast, and then binarized. The contour edge of the marker is extracted from the binarized image through morphological opening operation, connected domain analysis and other steps. The original sample image is contour aligned and size normalized according to the edge information, and finally a set of labeled images is generated according to the preset rules. Taking the foundation of the power tower as an example, the labeled image of the corresponding stage may be the contour line of the steel cage skeleton. With the labeled sample image set, it can be used to train the target detection model. The scheme adopts the Faster R-CNN algorithm, inputs the labeled image set into the algorithm, performs iterative training, and finally obtains the progress recognition model for power infrastructure projects. The model can accurately identify various types of construction markers in the monitoring video image and associate them with the known construction stage. In actual application, it only needs to collect video stream data from multiple monitoring cameras in real time, extract the target frame image after alignment and synchronization, and input it into the trained progress recognition model, and then the current construction stage data, including the stage number and the corresponding timestamp, can be automatically output. Taking the construction of the foundation of an electric power tower as an example, if a steel cage object is detected in the video, it can be determined that the current stage is the steel bar binding stage of the foundation construction. With the current stage data, the list of completed construction tasks can be accurately determined based on the pre-established construction plan mapping relationship, thereby calculating the real-time progress percentage of the project. At the same time, combined with the mapping relationship between construction tasks and cost expenditures, the estimated value of the overall project funding can be predicted. Taking the electric power tower as an example, if the foundation pouring is completed, it can be inferred that the project progress at this time is about 70%, and the capital consumption can be estimated based on the budget cost model. Finally, the program inputs the real-time project progress percentage and capital estimation data into the BIM (Building Information Model) three-dimensional model construction system to generate a digital twin model of the project.The model is processed through rendering optimization, lighting simulation, perspective setting, animation, etc. to form an intuitive set of visual elements, and the model is displayed through the data display terminal. Taking the power tower as an example, its BIM digital twin model will truly reflect the current progress of the tower construction, such as the completion of the foundation, pole, string arm level and other parts, providing all-round support for project management.

[0022] By executing the above steps, by collecting marker samples for multiple construction stages, and inputting the samples into the threshold segmentation algorithm for image annotation, a labeled image set is obtained, and then the Faster R-CNN algorithm is trained using the labeled image set to obtain a progress recognition model for power infrastructure projects. The progress recognition of real-time monitoring video images of the construction site can be accurately and efficiently performed, and the current construction stage data can be automatically output. This progress monitoring method based on artificial intelligence visual recognition technology does not require manual on-site inspections and records, greatly improves monitoring efficiency, reduces human operational errors, and avoids safety hazards. According to the identified current stage data, the real-time progress percentage of the project is accurately calculated through association mapping with the construction plan, and the project capital expenditure is estimated based on the progress percentage, which provides a basis for project managers to timely understand the use of funds and adjust the capital allocation strategy, thereby achieving more refined and rational capital management and control. Combine the project progress data with BIM (Building Information Model) technology, build a BIM digital twin model of the project based on the progress percentage and funding estimate, and perform various forms of visualization on the model, including material rendering, lighting simulation, perspective setting, and animation, and finally output a full range of visual model element sets, which are intuitively displayed using a data display terminal. This BIM digital twin modeling technology enables project managers to clearly understand the implementation progress and funding consumption of the entire project, providing a scientific reference for subsequent decision-making.

[0023] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Divide multiple key areas in the preset historical construction site to obtain a key area set, where each key area in the key area set corresponds to a construction stage; (2) Based on the key area set, historical video is collected for each key area to obtain a historical monitoring video stream set; (3) parsing the historical surveillance video stream set frame by frame to obtain a key frame image set, wherein the key frame image set includes a key frame image of each key area; (4) Performing image enhancement processing on the key frame image set to obtain an enhanced image set; (5) Perform image segmentation on the enhanced image set, extract the contour of the construction object in each image, and obtain a set of construction object contours; (6) Perform template matching on the construction object contour set according to the preset construction stage marker template to obtain a matching result set; (7) Based on the matching result set, extract the construction object contour with the highest matching degree with the marker template to obtain a matching contour set; (8) Normalizing the size of the matching contour set to obtain a normalized marker sample; (9) Encode the normalized marker samples according to the sequence of the construction stages to obtain a coded marker sample set; (10) Classify the coded marker sample set according to the construction stage to obtain a marker sample set, wherein the marker sample set includes marker samples of each construction stage.

[0024] Specifically, for the processing of historical construction site video data, it is necessary to first divide the site into multiple key areas according to the characteristics of the project. Each area corresponds to a major construction stage. For example, taking the construction of a transmission line tower as an example, the entire site can be divided into a foundation excavation area, a main body erection area, an insulator installation area, etc. Through this division, a key area set can be obtained, and each area in the set actually corresponds to a certain construction stage of the project. According to the key area set, historical video collection will be performed on each area to obtain a complete monitoring video stream data set covering each stage. Taking the tower construction as an example, monitoring videos of areas such as the foundation excavation area and the main body erection area will be collected to form a multi-channel video stream set. After obtaining the video stream set, it will be parsed frame by frame to extract key frame images from it. The so-called key frame refers to a representative picture that can reflect the main content of the current construction stage. Through this step, a key frame image set can be obtained, which contains the key frame images of each key area corresponding to the stage. In order to improve the effect of subsequent image processing, the key frame image set will be subjected to image enhancement processing, such as enhancing contrast and sharpening edges, to obtain an enhanced image set, which is prepared for subsequent image segmentation. Next, the image segmentation algorithm will be applied to the enhanced image set to automatically extract the main construction object contours in each image to form a construction object contour set. Taking the iron tower construction as an example, after segmentation, the contours of the excavator in the foundation excavation area and the tower in the main erection area may be obtained. After the construction object contour set is obtained, the contour set will be template matched according to the pre-set marker templates of each stage to obtain a matching result set. The template can be a labeled sample image used to match and identify a specific construction object. For example, in the foundation excavation stage, the template can be a contour sample of an excavator. According to the matching result set, a group of construction object contours with the highest matching degree with the marker template will be extracted as the matching contour set of this stage. At the same time, in order to eliminate the influence of image size differences, the matching contour set will be size normalized to obtain a normalized marker sample set. Finally, the normalized marker samples will be encoded according to the order of the construction stages, and different labels will be assigned to form an encoded marker sample set. Then, the encoded sample set will be classified and summarized according to the construction stage, and a sample set containing marker samples of each stage can be finally obtained. Taking the construction of transmission line towers as an example, the system may extract the outline of the excavator from the video of the foundation excavation area as the marker sample of this stage; extract the outline of the tower from the main erection area as the sample of the main erection stage; extract the outline of the insulator from the insulator installation area as the sample of the corresponding stage, and classify and encode them in order of stage, finally forming a complete set of marker samples.

[0025] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) grayscale the marker sample set to obtain a grayscale marker sample set; (2) Performing histogram equalization on the grayscale marker sample set to obtain a grayscale marker sample set with enhanced contrast; (3) According to a preset threshold, the contrast-enhanced grayscale marker sample set is binarized to obtain a binary marker sample set; (4) Performing a morphological opening operation on the binary marker sample set to obtain a binary marker sample set after the opening operation; (5) Performing a connected domain analysis on the binary marker sample set after the opening operation to obtain a binary marker sample set with connected domain labels; (6) According to a preset area threshold, the binary marker sample set with connected domain labels is filtered to obtain a filtered binary marker sample set; (7) Extracting edges from the filtered binary marker sample set to obtain a marker contour edge set; (8) According to the marker contour edge set, the original marker samples are contour aligned to obtain a marker sample set after contour alignment; (9) Normalizing the size of the marker sample set after contour alignment to obtain a size-normalized marker sample set; (10) According to preset annotation rules, the size-normalized marker sample set is annotated to obtain an annotated image set.

[0026] Specifically, the marker sample set will be grayed to convert the original color image into a gray image. This step can simplify the subsequent image analysis calculation. Taking the construction of the transmission line tower as an example, the color sample image of the excavator in the foundation excavation stage will be converted into a gray image. Next, the grayed sample set is subjected to histogram equalization to enhance the contrast and clarity of the image. Histogram equalization is a commonly used image enhancement technology. It adjusts the grayscale histogram distribution of the image so that the pixel values ​​are evenly distributed in the entire grayscale range, thereby improving the contrast and detail information of the image. This step is particularly important for the complex environment of the power engineering site. After the contrast-enhanced grayscale sample image is obtained, it will be binarized according to a pre-set threshold. Binarization is to reduce the image pixel value to only two values, 0 (black) or 1 (white). This process can filter out unnecessary detail information in the image and highlight the target object of interest. Taking the tower construction as an example, only the outline shape of the excavator and other equipment will remain on the binarized image. In order to further optimize the binary image, eliminate noise and make the target object more continuous and complete, the morphological opening operation will be performed on the binary sample set. The opening operation is a commonly used morphological operation in preprocessing. It first performs an erosion operation to remove small area noise, and then performs an expansion operation to restore the eroded target area, thereby filtering out isolated noise points and smoothing the target contour. Next, the binary sample set processed by the opening operation will be subjected to a connected domain analysis. Connected domain analysis can group pixels in an image according to connectivity, thereby separating different connected areas and assigning different labels to each connected area. This step helps to filter out noise areas based on area size in the subsequent process. According to the pre-set area threshold, the binary sample set with connected domain labels will be filtered to remove connected areas with an area less than the threshold, and retain the main area with a larger area, thereby obtaining a filtered binary marker sample set. Taking the power engineering as an example, at this time, only the contours of equipment such as excavators and towers that occupy a larger area will remain on the image. With the filtered binary samples, an edge extraction operation will be performed to obtain the marker contour edge set. The edge is the location where the pixel value in the image changes dramatically, which is crucial for extracting the contour of the object. Commonly used edge detection operators include Canny, Sobel, etc. Next, the contour of the original marker sample image will be aligned based on the extracted marker contour edge set. This step is to eliminate the contour displacement and deformation caused by factors such as viewing angle and scale, so as to obtain a normalized contour. Finally, the marker sample set after contour alignment will be size normalized to keep the size of all sample images consistent, in preparation for subsequent feature extraction and model training. At the same time, the normalized sample set will be annotated according to the pre-set annotation rules to generate a labeled image set as the input data for the training progress recognition model.Taking the transmission line tower project as an example, the annotated image set may include the excavator outline image in the foundation excavation stage, the tower outline image in the main erection stage, etc. Each image is labeled with the corresponding construction stage to guide the model to learn and recognize the feature patterns of different stages.

[0027] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Numbering multiple surveillance cameras within a preset target area to obtain a camera number set, where each camera corresponds to a unique number; (2) According to the camera number set, the video stream of each camera is collected in real time to obtain a set of multi-channel monitoring video streams; (3) Synchronize and align the multi-channel surveillance video stream set to obtain a synchronized surveillance video stream set; (4) extracting target frames from the synchronized surveillance video stream set at a preset time interval to obtain a target frame image set, where each target frame image corresponds to a timestamp; (5) performing image preprocessing on the target frame image set to obtain a preprocessed target frame image set; (6) Inputting the preprocessed target frame image set into the electric power infrastructure project progress recognition model, performing feature extraction and target detection on each target frame image, and obtaining a target frame detection result set; (7) Perform non-maximum suppression and bounding box merging processing on the target frame detection result set to obtain an optimized target frame detection result set; (8) Based on the optimized target frame detection result set, the construction stage of each target frame image is identified to obtain the target frame construction stage identification result set; (9) Sort the target frame construction phase recognition result set according to the timestamp to obtain a time-series construction phase recognition result sequence; (10) Voting fusion is performed on the time-series construction phase identification result sequence, and the construction phase corresponding to each timestamp is determined according to the majority voting principle to obtain the current phase data, including the current construction phase number and the corresponding timestamp.

[0028] Specifically, it is necessary to number multiple surveillance cameras pre-deployed at the target construction site, assign a unique number identifier to each camera, and thus obtain a camera number set. Taking the construction of a transmission line tower as an example, it may be necessary to deploy multiple surveillance cameras in key locations such as the foundation excavation area, the tower pole area, and the insulator installation area. With the camera number set, video stream data will be collected from each camera in real time according to this set to form a surveillance video stream set containing multiple video sources. It is worth noting that for subsequent data processing, these video streams need to be accurately time-synchronized to ensure that each video is completely synchronized in time. Next, key frames will be extracted from the synchronized multi-channel surveillance video stream set at a preset time interval to construct a target frame image set. Each extracted target frame image will be automatically associated with a timestamp information to record the time point corresponding to the frame image. The purpose of this step is to reduce data redundancy and improve subsequent processing efficiency. In order to eliminate noise in the target frame image and enhance the quality, it is also necessary to perform image preprocessing operations on the target frame image set, such as denoising, enhanced sharpening, etc., to obtain a preprocessed high-quality target frame image set. After the preprocessing is completed, the high-quality target frame image set will be input into the previously trained power infrastructure project progress recognition model. The model will perform feature extraction and target detection operations on each frame image, automatically identify the marker targets related to each construction stage in the image, and generate corresponding bounding boxes and scores for each detected target, thereby obtaining a target frame detection result set. Since the detection results of the model may have certain redundancy, such as detecting multiple overlapping bounding boxes for the same target, it is necessary to perform non-maximum suppression and bounding box merging processing on the target frame detection result set, remove redundant detection boxes, retain the box with the highest detection score, and obtain an optimized detection result set. With the optimized detection results, the corresponding construction stage can be identified for each frame image according to the detected target type, and the recognition results can be summarized into a target frame construction stage recognition result set. Taking the tower construction as an example, if the excavator target appears in the detection result, it can be judged that the current stage is the foundation excavation stage. Next, the construction stage recognition result set will be sorted according to the timestamp order corresponding to each frame image, so as to obtain a time-series construction stage recognition result sequence unfolded by timeline. This step is to capture the dynamic change trajectory of the construction progress. Finally, for the same time point, different cameras may produce different stage recognition results. To solve this problem, the scheme proposes a voting fusion processing strategy. Specifically, multiple stage recognition results of the same timestamp will be voted, and the final construction stage number of the time point will be determined according to the majority voting principle, so as to output the current stage data, including the construction stage number and the corresponding timestamp.

[0029] In a specific embodiment, the process of performing image preprocessing on the target frame image set to obtain the preprocessed target frame image set may specifically include the following steps: (1) Perform image denoising on the target frame image set by using a median filtering algorithm to obtain a denoised target frame image set; (2) performing image enhancement on the denoised target frame image set to obtain an enhanced target frame image set; (3) performing edge-preserving filtering on the enhanced target frame image set using a bilateral filtering algorithm to obtain an edge-enhanced target frame image set; (4) Performing color space conversion on the target frame image set with edge enhancement to obtain the target frame image set in the HSV color space; (5) Extract channel data of the target frame image set in the HSV color space to obtain a hue channel image set, a saturation channel image set, and a lightness channel image set; (6) Performing histogram statistics on the hue channel image set to obtain a hue histogram feature vector set, performing histogram statistics on the saturation channel image set to obtain a saturation histogram feature vector set, and performing histogram statistics on the lightness channel image set to obtain a lightness histogram feature vector set; (7) Concatenate the hue histogram feature vector set, the saturation histogram feature vector set, and the lightness histogram feature vector set to obtain the HSV color histogram feature vector set; (8) Normalizing the HSV color histogram feature vector set to obtain a normalized HSV color histogram feature vector set; (9) The normalized HSV color histogram feature vector set is used as the preprocessed target frame image set.

[0030] Specifically, due to the complex on-site environment, the acquired surveillance video may be interfered by noise, so it is necessary to denoise the target frame image set. The scheme adopts the median filtering algorithm, which effectively eliminates the salt and pepper noise and speckle noise in the image by replacing the gray value of the pixel point with the median of the gray value of the pixel point neighborhood, and obtains a clear image set after denoising. Next, in order to enhance the clarity and contrast of the image, the scheme performs image enhancement on the denoised target frame image set. Common methods of image enhancement include histogram equalization, gamma correction, Laplace sharpening, etc. Taking the transmission line project as an example, for videos shot in outdoor environments, it is often necessary to enhance the contrast so that the target object is more separated from the background. With the enhanced target frame image set, the scheme continues to use the bilateral filtering algorithm to filter it. Bilateral filtering is a very effective image filtering technology that can both eliminate noise and retain edge details. It replaces the central pixel value by calculating the weighted average of the gray value similarity and spatial distance similarity of the neighborhood pixels, thereby achieving an edge-preserving smoothing filtering effect and obtaining an edge-enhanced target frame image set. For the construction site environment, it is crucial to be able to more clearly distinguish the edges of the equipment contours. Next, the scheme converts the edge-enhanced target frame image set from the RGB color space to the HSV color space. The HSV color space can better describe the properties of color, such as hue, saturation, and brightness, and is therefore more suitable for subsequent color feature extraction. After the conversion, the target frame image set in the HSV color space can be obtained. Then, the scheme extracts channel data from the image set in the HSV color space to obtain three channel image sets of hue channel, saturation channel, and brightness channel. By separating different color channels, the required color feature information can be extracted more efficiently. Next, for each channel image set, the scheme uses histogram statistics to extract the three color histogram features of hue, saturation, and brightness, thereby obtaining a hue histogram feature vector set, a saturation histogram feature vector set, and a brightness histogram feature vector set. These three feature vector sets can fully describe the color distribution of the image. By splicing the above three feature vector sets, a complete HSV color histogram feature vector set can be obtained to characterize the overall color characteristics of the target frame image. Finally, the HSV color histogram feature vector set is normalized and mapped to a uniform numerical range, thereby obtaining a normalized HSV color histogram feature vector set. This normalized feature vector set is regarded as a representation of the preprocessed target frame image and will be input into the electric power infrastructure project progress recognition model for feature extraction and target detection. Taking the transmission line tower project as an example, for the scene images captured by the surveillance video, this preprocessing pipeline can effectively remove noise, enhance contrast and edge details, and extract distinguishable color features, laying the foundation for subsequent target detection of towers, excavators and other equipment.

[0031] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Establish a mapping relationship database between construction stages and construction tasks, where each construction stage corresponds to multiple specific construction tasks, and obtain a construction stage-task mapping table; (2) Based on the current stage data, search the completed construction task list corresponding to the current construction stage in the construction stage-task mapping table to obtain the completed task list; (3) For each task in the completed task list, obtain the corresponding planned engineering quantity from the pre-established construction task-engineering quantity mapping table, and count the cumulative completed engineering quantities of all completed tasks to obtain the cumulative completed engineering quantities; (4) Obtain the total planned project quantity from the construction task-project quantity mapping table, divide the cumulative completed project quantity by the total planned project quantity, and obtain the current project progress percentage; (5) Establish a mapping relationship database between construction tasks and cost expenditures to obtain a construction task-cost mapping table; (6) Perform cost expenditure analysis based on the completed task list and the construction task-cost mapping table to obtain the cumulative actual cost expenditure of the completed tasks; (7) Analyze the budget cost expenditure of the completed task list based on the current project progress percentage to obtain the budget cost expenditure of the completed tasks; (8) Calculate the difference between the accumulated actual cost expenditure of the completed tasks and the budgeted cost expenditure of the completed tasks to obtain the difference data; (9) Based on the difference data and the current percentage of project progress, the project capital expenditure is estimated to obtain the estimated value of the project capital expenditure.

[0032] Specifically, a mapping relationship library between construction stages and specific construction tasks is established and stored in the form of a construction stage-task mapping table. In this mapping table, each construction stage corresponds to multiple specific construction tasks. Taking the transmission line tower project as an example, the foundation excavation stage may correspond to tasks such as excavating the foundation, making templates, and tying bottom hoop steel bars; the main erection stage corresponds to tasks such as assembling the tower body and lifting it into place. With this mapping table, the corresponding completed construction task list can be found in the table based on the current stage data obtained from the visual recognition link. If the recognition result shows that it is currently in the foundation excavation stage, then the completed task list may include excavating the foundation and making templates. Next, for each task in the completed task list, the planned engineering quantity value corresponding to the task will be obtained from the pre-established construction task-engineering quantity mapping table, and the engineering quantities of all completed tasks will be accumulated and counted to obtain the cumulative completed engineering quantity of the entire project so far. At the same time, it is also necessary to obtain the total planned engineering quantity of the entire project from the construction task-engineering quantity mapping table. The actual progress percentage of the current project can be calculated by dividing the cumulative completed engineering quantity by the total planned engineering quantity. On the other hand, in order to predict the use of project funds, the solution also establishes a mapping relationship library between construction tasks and cost expenditures, which is stored in the form of a construction task-cost mapping table. According to the list of completed tasks and this mapping table, the cumulative actual cost expenditures corresponding to the completed tasks can be analyzed and counted. Similarly, the budget cost expenditure analysis of the completed task list will be performed according to the progress percentage of the current project to obtain the budget cost expenditure values ​​of the completed tasks. Next, the cumulative actual cost expenditures of the completed tasks will be compared with the budget cost expenditures to calculate the difference data between the two. This difference data can reflect the degree of deviation between the actual expenditure and the budget. Finally, based on the difference data and the current project progress percentage, the solution will use a certain prediction algorithm or empirical model to estimate the capital expenditure of the entire project and obtain the final estimated value of the project capital expenditure. Taking the transmission line tower project as an example, assume that the current visual recognition results show that the main erection stage has been completed, and according to the engineering quantity accounting, the project progress has reached 60%. At the same time, according to the task-cost mapping table analysis, the actual expenditure of the completed tasks is 50 million yuan, while the progress budget should be 48 million yuan, a difference of 2 million yuan. On this basis, the plan may combine industry experience data to predict that the funding demand for the subsequent 40% of the project volume is 64 million yuan, so the final funding expenditure estimate for the entire project is 5000+6400=114 million yuan.

[0033] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Decompose the construction plan into vectors according to the current project progress percentage to obtain a set of engineering quantity feature vectors; (2) Perform feature mapping on the parameterized algorithm according to the engineering quantity feature vector set to obtain a resource consumption mapping set; (3) Cost accounting is performed on the resource consumption mapping set to obtain cost budget data; (4) Correct and adjust the estimated value of project capital expenditure based on the cost budget data to obtain the corrected estimated value of project capital expenditure; (5) Input the revised estimated value of project capital expenditure into the BIM 3D model construction algorithm to obtain the original BIM digital twin model; (6) Render the original BIM digital twin model with materials to obtain a BIM model after material rendering; (7) Perform lighting simulation based on the BIM model after material rendering to obtain the BIM model after lighting rendering; (8) Setting the viewing angle of the BIM model after lighting rendering to obtain a BIM visualization model after the viewing angle setting; (9) Animating the BIM visualization model after the perspective is set to obtain an animated BIM visualization model; (10) The animated BIM visualization model is output as a visualization element set for model display on a preset data display terminal.

[0034] It should be noted that the construction plan will be vector decomposed according to the calculated percentage of the current project progress. The so-called vector decomposition refers to taking the total progress of the project as a vector and decomposing it according to the workload of the specific construction task to obtain a set of engineering quantity feature vectors, in which each vector component corresponds to the proportion of the engineering quantity of a construction task. Taking the transmission line tower as an example, if the current progress is 50%, it may be decomposed into task quantity components such as 20% foundation excavation and 30% main body erection. Next, this engineering quantity feature vector set will be input into a parameterized algorithm for feature mapping. The so-called parameterized algorithm refers to mapping the feature of the engineering quantity to the actual resource consumption and calculating it according to the pre-modeled parameter relationship function. Through this step, a resource consumption mapping set can be obtained, in which each component corresponds to the amount of resources required for different construction tasks, such as labor, materials, mechanical equipment, etc. After having the resource consumption mapping set, cost accounting operations will be performed on it, and each resource consumption will be converted into actual monetary cost data, thereby obtaining the cost budget data set for each task of the entire project. This process requires calling the pre-established resource pricing parameter model. Next, the previously estimated project funding expenditure value will be revised and adjusted based on the cost budget data obtained in the previous step to obtain a revised and more accurate project funding expenditure estimate. This estimate will be used as the main parameter input for building the BIM digital twin model. The revised funding estimate is input into the BIM three-dimensional model construction algorithm to generate the original BIM digital twin model of the project. In the initial state of this three-dimensional model, it only contains basic geometric shapes and spatial layout information. In order to make the BIM model more realistic and intuitive, a series of rendering and optimization operations need to be performed on it. The first is material rendering, which is to give real material properties to each component in the model, such as reinforced concrete, insulating glass, etc., and adjust fine texture details to obtain a BIM model after material rendering. Secondly, the lighting simulation will be performed based on the rendered model to simulate real optical effects such as light refraction and shadow projection, so that the model is more three-dimensional and has higher glossiness visually, and a BIM model after lighting rendering is obtained. Next, the model needs to be set to a reasonable viewing angle, scaling and other parameters to ensure that the key parts of the model are clear at a glance. Through this step, the BIM visualization model after the viewing angle setting can be obtained. Finally, the solution will also animate the visualization model, giving it dynamic characteristics in the time dimension, simulating the specific process of the entire project construction, such as foundation pouring, pole assembly, insulator assembly, etc., to form an animated BIM visualization model. Taking the transmission line project as an example, the animation model can truly reproduce the entire process of tower construction. The packaged output of the animated BIM model can form a complete set of visualization elements, which includes different perspectives, materials, lighting and animation content of the model.Ultimately, the set of visualization elements will be transmitted to preset data display terminals, such as large-screen splicing displays, virtual reality devices, etc., to provide relevant personnel with an intuitive, vivid, and comprehensive engineering display experience.

[0035] See also Figure 2 , Figure 2 A power infrastructure project monitoring system based on deep learning is provided in an embodiment of the present invention. Figure 2 As shown, the system includes: The collection module 201 is used to collect stage marker samples for multiple construction stages to obtain a marker sample set, wherein the marker sample set includes marker samples for each construction stage; Annotation module 202, used for inputting the marker sample set into a threshold segmentation algorithm for image annotation to obtain an annotated image set; A training module 203 is used to perform model training on a preset Faster R-CNN algorithm using the annotated image set to obtain a power infrastructure project progress recognition model; The recognition module 204 is used to collect monitoring video images of a preset target area in real time, input the monitoring video images into the electric power infrastructure project progress recognition model for progress recognition, and obtain current stage data; A mapping module 205 is used to perform construction scheme association mapping according to the current stage data, obtain the current project progress percentage, and predict the estimated value of project capital expenditure according to the current project progress percentage; The display module 206 is used to construct a BIM digital twin model according to the current engineering progress percentage and the estimated value of the project capital expenditure, construct visualization elements for the BIM digital twin model, obtain a visualization element set, and display the model on a preset data display terminal through the visualization element set.

[0036] Through the collaborative work of the above modules, by collecting marker samples for multiple construction stages, and inputting the samples into the threshold segmentation algorithm for image annotation, a set of annotated images is obtained, and then the Faster R-CNN algorithm is trained using the annotated image set to obtain a progress recognition model for power infrastructure projects. The real-time monitoring video images of the construction site can be accurately and efficiently identified, and the current construction stage data can be automatically output. This progress monitoring method based on artificial intelligence visual recognition technology does not require manual on-site inspections and records, greatly improves monitoring efficiency, reduces human operational errors, and avoids safety hazards. According to the identified current stage data, the real-time progress percentage of the project is accurately calculated through association mapping with the construction plan, and the project capital expenditure is estimated based on the progress percentage, which provides a basis for project managers to timely understand the use of funds and adjust the capital allocation strategy, thereby achieving more refined and rational capital management and control. Combine the project progress data with BIM (Building Information Model) technology, build a BIM digital twin model of the project based on the progress percentage and funding estimate, and perform various forms of visualization on the model, including material rendering, lighting simulation, perspective setting, and animation, and finally output a full range of visual model element sets, which are intuitively displayed using a data display terminal. This BIM digital twin modeling technology enables project managers to clearly understand the implementation progress and funding consumption of the entire project, providing a scientific reference for subsequent decision-making.

[0037] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the embodiments, a person skilled in the art should understand that the specific implementation modes of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A power infrastructure project monitoring method based on deep learning, characterized in that: include: Collecting stage marker samples for multiple construction stages to obtain a marker sample set, wherein the marker sample set includes marker samples for each construction stage; Inputting the marker sample set into a threshold segmentation algorithm for image annotation to obtain an annotated image set; The preset Faster R-CNN algorithm is trained by using the annotated image set to obtain a power infrastructure project progress recognition model; Collecting surveillance video images of a preset target area in real time, inputting the surveillance video images into the electric power infrastructure project progress recognition model for progress recognition, and obtaining current stage data; Perform construction scheme association mapping based on the current stage data to obtain the current project progress percentage, and predict the estimated value of project capital expenditure based on the current project progress percentage; A BIM digital twin model is constructed according to the current engineering progress percentage and the estimated value of the project capital expenditure, visualization elements are constructed for the BIM digital twin model to obtain a visualization element set, and the model is displayed on a preset data display terminal through the visualization element set.

2. The power infrastructure project monitoring method based on deep learning according to claim 1 is characterized in that: The step of collecting stage marker samples for multiple construction stages to obtain a marker sample set, wherein the marker sample set includes marker samples for each construction stage, includes: Dividing a plurality of key areas in a preset historical construction site to obtain a key area set, wherein each key area in the key area set corresponds to a construction stage; According to the key area set, historical video collection is performed on each key area to obtain a historical monitoring video stream set; Parsing the historical surveillance video stream set frame by frame to obtain a key frame image set, wherein the key frame image set includes a key frame image of each key area; Performing image enhancement processing on the key frame image set to obtain an enhanced image set; Performing image segmentation on the enhanced image set, extracting the outline of the construction object in each image, and obtaining a construction object outline set; According to the preset construction stage marker template, template matching is performed on the construction object contour set to obtain a matching result set; According to the matching result set, extracting the construction object contour with the highest matching degree with the marker template to obtain a matching contour set; Performing size normalization processing on the matching contour set to obtain a normalized marker sample; According to the sequence of the construction phases, the normalized marker samples are encoded to obtain a coded marker sample set; The coded marker sample set is classified according to the construction stage to obtain the marker sample set, wherein the marker sample set includes marker samples of each construction stage.

3. The power infrastructure project monitoring method based on deep learning according to claim 1 is characterized in that: The step of inputting the marker sample set into a threshold segmentation algorithm for image annotation to obtain an annotated image set comprises: Performing grayscale processing on the marker sample set to obtain a grayscale marker sample set; Performing histogram equalization on the grayscale marker sample set to obtain a grayscale marker sample set with enhanced contrast; According to a preset threshold, binarization processing is performed on the contrast-enhanced grayscale marker sample set to obtain a binarized marker sample set; Performing a morphological opening operation on the binary marker sample set to obtain a binary marker sample set after the opening operation; Performing a connected domain analysis on the binary marker sample set after the opening operation to obtain a binary marker sample set with connected domain labels; According to a preset area threshold, the binary marker sample set with connected domain marks is filtered to obtain a filtered binary marker sample set; Performing edge extraction on the filtered binary marker sample set to obtain a marker contour edge set; According to the marker contour edge set, contour alignment is performed on the original marker samples to obtain a marker sample set after contour alignment; Normalizing the size of the marker sample set after the contour alignment to obtain a size-normalized marker sample set; According to preset annotation rules, image annotation is performed on the size-normalized marker sample set to obtain an annotated image set.

4. The power infrastructure project monitoring method based on deep learning according to claim 1 is characterized in that: The step of collecting monitoring video images of a preset target area in real time, inputting the monitoring video images into the electric power infrastructure project progress recognition model for progress recognition, and obtaining current stage data includes: Numbering multiple surveillance cameras in a preset target area to obtain a camera number set, wherein each camera corresponds to a unique number; According to the camera number set, the video stream of each camera is collected in real time to obtain a multi-channel monitoring video stream set; Synchronizing and aligning the multi-channel monitoring video stream set to obtain a synchronized monitoring video stream set; Extracting target frames from the synchronized surveillance video stream set at preset time intervals to obtain a target frame image set, each target frame image corresponding to a timestamp; Performing image preprocessing on the target frame image set to obtain a preprocessed target frame image set; Inputting the preprocessed target frame image set into the electric power infrastructure project progress recognition model, performing feature extraction and target detection on each target frame image, and obtaining a target frame detection result set; Performing non-maximum suppression and bounding box merging processing on the target frame detection result set to obtain an optimized target frame detection result set; According to the optimized target frame detection result set, the construction stage of each target frame image is identified to obtain a target frame construction stage identification result set; Sorting the target frame construction phase recognition result set according to timestamps to obtain a time-series construction phase recognition result sequence; The time-series construction phase identification result sequence is voted and fused, and the construction phase corresponding to each timestamp is determined according to the majority voting principle to obtain the current phase data, including the current construction phase number and the corresponding timestamp.

5. The power infrastructure project monitoring method based on deep learning according to claim 4 is characterized in that: The step of performing image preprocessing on the target frame image set to obtain the preprocessed target frame image set comprises: Performing image denoising on the target frame image set by using a median filtering algorithm to obtain a denoised target frame image set; Performing image enhancement on the denoised target frame image set to obtain an enhanced target frame image set; Performing edge-preserving filtering on the enhanced target frame image set by using a bilateral filtering algorithm to obtain an edge-enhanced target frame image set; Performing color space conversion on the edge enhanced target frame image set to obtain a target frame image set in an HSV color space; Extracting channel data of the target frame image set in the HSV color space to obtain a hue channel image set, a saturation channel image set, and a lightness channel image set; Performing histogram statistics on the hue channel image set to obtain a hue histogram feature vector set, performing histogram statistics on the saturation channel image set to obtain a saturation histogram feature vector set, and performing histogram statistics on the lightness channel image set to obtain a lightness histogram feature vector set; The hue histogram feature vector set, the saturation histogram feature vector set and the lightness histogram feature vector set are concatenated to obtain an HSV color histogram feature vector set; Normalizing the HSV color histogram feature vector set to obtain a normalized HSV color histogram feature vector set; The normalized HSV color histogram feature vector set is used as the preprocessed target frame image set.

6. The power infrastructure project monitoring method based on deep learning according to claim 1 is characterized in that: The step of performing construction scheme association mapping according to the current stage data to obtain the current engineering progress percentage, and predicting the estimated value of project capital expenditure according to the current engineering progress percentage includes: Establish a mapping relationship library between construction stages and construction tasks, where each construction stage corresponds to multiple specific construction tasks, and obtain a construction stage-task mapping table; According to the current stage data, the completed construction task list corresponding to the current construction stage is searched in the construction stage-task mapping table to obtain the completed task list; For each task in the completed task list, the corresponding planned engineering quantity is obtained from a pre-established construction task-engineering quantity mapping table, and the cumulative completed engineering quantity of all completed tasks is counted to obtain the cumulative completed engineering quantity; Obtain the total planned engineering quantity from the construction task-engineering quantity mapping table, divide the cumulative completed engineering quantity by the total planned engineering quantity, and obtain the current engineering progress percentage; Establish a mapping relationship database between construction tasks and cost expenditures to obtain a construction task-cost mapping table; Perform cost expenditure analysis based on the completed task list and the construction task-cost mapping table to obtain the accumulated actual cost expenditure of the completed tasks; Performing a budget cost expenditure analysis on the completed task list according to the current engineering progress percentage to obtain the budget cost expenditure of the completed tasks; Calculate the difference between the accumulated actual cost expenditure of the completed tasks and the budgeted cost expenditure of the completed tasks to obtain difference data; The project capital expenditure estimate is predicted based on the difference data and the current engineering progress percentage to obtain the project capital expenditure estimate.

7. The power infrastructure project monitoring method based on deep learning according to claim 6 is characterized in that: The step of constructing a BIM digital twin model according to the current engineering progress percentage and the estimated value of the project capital expenditure, constructing visualization elements for the BIM digital twin model to obtain a visualization element set, and performing a model display step on a preset data display terminal through the visualization element set includes: Performing vector decomposition on the construction plan according to the current engineering progress percentage to obtain a set of engineering quantity feature vectors; Performing feature mapping on the parameterized algorithm according to the engineering quantity feature vector set to obtain a resource consumption mapping set; Performing cost accounting on the resource consumption mapping set to obtain cost budget data; Correcting and adjusting the estimated value of project capital expenditure according to the cost budget data to obtain a corrected estimated value of project capital expenditure; Inputting the revised estimated value of project capital expenditure into the BIM three-dimensional model construction algorithm to obtain the original BIM digital twin model; Performing material rendering on the original BIM digital twin model to obtain a BIM model after material rendering; Performing lighting simulation according to the BIM model rendered with the material to obtain a BIM model rendered with lighting; Setting a viewing angle for the BIM model after illumination rendering to obtain a BIM visualization model after the viewing angle setting; Animating the BIM visualization model after the viewing angle is set to obtain an animated BIM visualization model; The animated BIM visualization model is output as a visualization element set for model display on a preset data display terminal.

8. A power infrastructure project monitoring system based on deep learning, used to execute the power infrastructure project monitoring method based on deep learning as claimed in any one of claims 1 to 7, characterized in that: include: A collection module, used for collecting stage marker samples for multiple construction stages to obtain a marker sample set, wherein the marker sample set includes marker samples for each construction stage; A labeling module, used for inputting the marker sample set into a threshold segmentation algorithm for image labeling to obtain a labeling image set; A training module, used to perform model training on a preset Faster R-CNN algorithm using the annotated image set to obtain a power infrastructure project progress recognition model; An identification module is used to collect monitoring video images of a preset target area in real time, input the monitoring video images into the electric power infrastructure project progress identification model for progress identification, and obtain current stage data; A mapping module is used to perform construction scheme association mapping according to the current stage data, obtain the current project progress percentage, and predict the estimated value of project capital expenditure according to the current project progress percentage; A display module is used to build a BIM digital twin model according to the current engineering progress percentage and the estimated value of the project capital expenditure, construct visualization elements for the BIM digital twin model to obtain a visualization element set, and display the model on a preset data display terminal through the visualization element set.

Images