A method and device for dynamic monitoring of architectural site defects based on crowdsourced imagery

By performing georeferencing and disease identification on crowdsourced images, a correspondence between images and site units is established, solving the problems of low efficiency in surveying architectural site diseases and high image acquisition costs in existing technologies, and realizing continuous monitoring and dynamic updating of architectural site diseases.

CN123071925APending Publication Date: 2026-11-10WUHAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611190937.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-06
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing surveys of architectural ruins suffer from problems such as low efficiency of manual inspections, high cost of professional image acquisition, difficulty in directly using crowdsourced images for continuous monitoring, and lack of dynamic updates to the results.

Method used

By regularly acquiring, georeferencing, identifying defects, detecting changes, and updating the image database of crowdsourced images, a correspondence between images and architectural heritage units is established, a defect identification model based on image visual content is constructed, and continuous monitoring and management of defect information are carried out.

Benefits of technology

It improves the flexibility and continuity of disease monitoring data acquisition without relying on specialized field collection and professional surveying and mapping images, enhances the spatial availability of crowdsourced images, provides clear image evidence to support disease identification results, and supports the temporal update of disease change records.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123071925A_ABST
    Figure CN123071925A_ABST
Patent Text Reader

Abstract

This invention discloses a method and apparatus for dynamic monitoring of architectural site defects based on crowdsourced imagery. The method acquires relevant crowdsourced images and their metadata for architectural sites at preset time intervals, and performs non-target scene filtering, duplicate image removal, and metadata standardization on candidate images to obtain a valid set of crowdsourced images. Based on the spatial association information carried by the valid crowdsourced images, spatial reference and site unit assignment are performed in conjunction with architectural site vector data. Defect identification is performed based on the visual content of the assigned images, extracting defect categories, defect locations or response areas, and identification confidence levels. Defect identification results for the same architectural site unit at different times are compared to generate a defect change record. The relevant data is then written into an architectural site defect image database for dynamic updates. This invention improves the efficiency of architectural site defect information acquisition, spatial organization, and continuous monitoring.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of digital protection of cultural heritage and geographic information processing technology, specifically relating to a method and device for dynamic monitoring of architectural site defects based on crowdsourced images. Background Technology

[0002] Architectural sites are an important component of cultural heritage. Exposed to the natural environment and human activities over long periods, they are prone to damage such as cracks, collapses, peeling, and vegetation erosion. These damages not only weaken the structural stability of the architectural sites but also affect the preservation and display of their historical information. Therefore, the continuous discovery, location, recording, and monitoring of changes in architectural site damage are of great significance for cultural heritage protection, the identification of key hidden dangers, and restoration decisions.

[0003] Current surveys of architectural ruins primarily rely on manual inspections, on-site photography, drone mapping, or professional remote sensing monitoring. While manual inspections provide relatively intuitive on-site information, they suffer from drawbacks such as high workload, low efficiency, long cycles, and insufficient continuity when dealing with large-scale, dispersed ruins. Drone imagery and professional mapping data offer high accuracy, but their acquisition typically requires specialized equipment, personnel, and flight conditions, making it difficult to update a large number of architectural ruins frequently over extended periods.

[0004] With the development of internet maps, open image platforms, heritage-themed websites, and public upload platforms, a large amount of crowdsourced images with time, location, or source information are continuously accumulating. Crowdsourced images are characterized by their wide range of sources, rapid updates, low acquisition costs, close-to-the-ground perspective, and ability to reflect local details, providing a new data source for monitoring architectural site defects. However, crowdsourced images also suffer from problems such as heterogeneous sources, unstable spatial locations, significant differences in shooting perspectives, inconsistent image quality, numerous duplicate images, and difficulty in directly associating them with specific site units. Without effective geographic references, defect extraction, and change detection mechanisms, crowdsourced images are difficult to directly use for dynamic monitoring of architectural site defects. Summary of the Invention

[0005] This invention addresses the problems in existing surveys of architectural ruins, such as low efficiency of manual inspections, high cost of professional image acquisition, difficulty in directly using crowdsourced images for continuous monitoring, and lack of dynamic updates to ruin results. It proposes a method and device for dynamic monitoring of architectural ruins based on crowdsourced images. This method achieves continuous monitoring and management of architectural ruins' ruin information through periodic acquisition of crowdsourced images, georeferencing, ruin identification, change detection, and image database updates.

[0006] According to one aspect of the present invention, a method for dynamic monitoring of architectural ruins based on crowdsourced images is provided, comprising: step S1, acquiring crowdsourced images related to architectural ruins and their metadata at preset time intervals, and performing non-target scene filtering, duplicate image removal, and metadata standardization processing on candidate images to obtain a set of effective crowdsourced images; step S2, based on the spatial association information carried by the effective crowdsourced images and combined with architectural ruin vector data, performing spatial reference and ruin unit attribution on the effective crowdsourced images, and establishing a correspondence between images and architectural ruin units; step S3, based on the attribution to architectural ruins... The visual content of the effective crowdsourced images of the unit is used for disease identification, and disease information in the architectural ruins is extracted to obtain the disease category, disease location or response area and identification confidence level; Step S4, based on the crowdsourced images and disease identification results of the same architectural ruins unit at different times, change detection is performed to determine the change type of the disease identification results and generate disease change records; Step S5, the original images, disease identification results, geographic reference information, time information, source information and disease change records are written into the architectural ruins disease image library, and the architectural ruins disease image library is dynamically updated according to the crowdsourced images acquired periodically in the future.

[0007] As a further technical solution, step S2 includes: for valid crowdsourced images with site unit numbers, reading the site unit number and matching the site unit number with the corresponding number field in the architectural site vector data to directly establish the association between the image and the corresponding architectural site unit; for valid crowdsourced images without site unit numbers but with geographic coordinates, calculating the spatial distance between the image coordinates and each site unit in the architectural site vector data, and assigning the image to the architectural site unit that is closest and meets the preset distance threshold; for valid crowdsourced images without site unit numbers and clear geographic coordinates but with route clues or place name clues, combining the route range, place name pointing information, and architectural site vector data for auxiliary interpretation to determine the corresponding architectural site unit.

[0008] As a further technical solution, step S3 includes: constructing a building site disease identification model based on image visual content, wherein the disease identification model includes a target detection model and an image classification model; using the target detection model to identify crack-type diseases in valid crowdsourced images, and outputting the location range, category, and confidence level of crack-type diseases; using the image classification model combined with a sliding window method to perform local disease identification of collapse, peeling, and vegetation erosion-type diseases in valid crowdsourced images; performing confidence level filtering, duplicate response merging, and result integration on the disease identification results in the same image to obtain the disease identification results in a single image; associating the disease category, disease location or response area, identification confidence level, and image number in a single image with the corresponding building site unit to form a building site unit-level disease identification result.

[0009] As a further technical solution, step S4 includes: organizing the crowdsourced images and their disease identification results acquired at different times according to the architectural site unit number and image time information; comparing the disease identification results at different times for the same architectural site unit, wherein the disease identification results include one or more of disease category, number of detected images, detection location and identification confidence; determining the change type of the disease identification results within the architectural site unit based on the comparison results, wherein the change type includes one of the following: change from not detected to detected, change from detected to not detected, increase in the number of detected images, decrease in the number of detected images, and basically consistent detection status, and generating a corresponding disease change record.

[0010] As a further technical solution, step S5 includes: establishing a building site disease image database, which includes original image records, site unit records, image-site unit association records, disease identification result records, disease change records, and data source records; writing the original image records, site unit records, image-site unit association records, disease identification result records, disease change records, and data source records into the building site disease image database; when new crowdsourced images are obtained, repeating steps S1 to S4, and adding the new original images, image-site unit association records, disease identification result records, and disease change records to the building site disease image database to achieve dynamic updating of the building site disease image database.

[0011] As a further technical solution, step S1 includes: according to a preset collection cycle, acquiring relevant images of architectural sites from heritage websites, open image platforms, street view or crowdsourced map platforms, outdoor trajectory platforms, or user upload terminals based on the site unit number, spatial location, site name, or route range in the architectural site vector data; extracting and recording the source information, time information, and spatial association information for spatial attribution of candidate images; wherein, the source information includes the image source platform, the time information includes at least one of the collection time and upload time, and the spatial association information includes at least one of the site unit number, geographic coordinates, route clues, and place name clues; filtering candidate images for non-target scenes, removing images that do not contain the architectural site itself or are unrelated to the architectural site monitoring object; performing duplicate image removal and metadata standardization processing on candidate images, removing duplicate or highly similar images, and standardizing and organizing the source information, time information, and spatial association information of candidate images to obtain an effective crowdsourced image set.

[0012] According to one aspect of the present invention, a dynamic monitoring device for architectural ruins based on crowdsourced images is provided, comprising: an image acquisition module, configured to acquire crowdsourced images related to architectural ruins and their metadata at preset time intervals, perform non-target scene filtering, duplicate image removal, and metadata standardization processing on candidate images to obtain a set of valid crowdsourced images; a geographic reference module, configured to perform spatial reference and ruins unit affiliation on the valid crowdsourced images based on the spatial association information carried by the valid crowdsourced images and in combination with architectural ruins vector data, and establish a correspondence between images and architectural ruins units; and a ruins identification module, configured to identify ruins based on the ruins unit affiliation. The system uses the visual content of valid crowdsourced images to identify defects, extracts defect information from architectural ruins, and outputs defect categories, defect locations or response areas, and recognition confidence levels. A change detection module is used to detect changes in crowdsourced images and defect identification results of the same architectural ruin unit at different times, determine the type of change in the defect identification results, and generate defect change records. A defect image library construction module is used to write the original images, defect identification results, geographic reference information, time information, source information, and defect change records into the architectural ruin defect image library, and dynamically updates the architectural ruin defect image library based on subsequently acquired crowdsourced images.

[0013] As a further technical solution, the geographic reference module reads the site unit number of valid crowdsourced images with site unit numbers and matches the site unit number with the corresponding number field in the architectural site vector data to directly establish the association between the image and the corresponding architectural site unit. For valid crowdsourced images without site unit numbers but with geographic coordinates, the module calculates the spatial distance between the image coordinates and each site unit in the architectural site vector data, and assigns the image to the architectural site unit that is closest to it and meets the preset distance threshold. For valid crowdsourced images without site unit numbers and clear geographic coordinates but with route clues or place name clues, the module uses the route range, place name information, and architectural site vector data for auxiliary interpretation to determine the corresponding architectural site unit.

[0014] As a further technical solution, the disease identification module includes a target detection model and an image classification model. The target detection model is used to identify crack-type diseases in valid crowdsourced images and output the location range, category, and confidence level of the crack-type diseases. The image classification model is used to perform local disease identification of collapse, peeling, and vegetation erosion diseases in valid crowdsourced images using a sliding window method. The disease identification module is also used to perform confidence level filtering, duplicate response merging, and result integration on the disease identification results in the same image to obtain the disease identification results in a single image. The disease category, disease location or response area, identification confidence level, and image number in a single image are associated with the corresponding architectural heritage unit to form architectural heritage unit-level disease identification results.

[0015] As a further technical solution, the change detection module organizes the crowdsourced images and their disease identification results acquired at different times according to the building site unit number and image time information. For the same building site unit, the module compares the disease identification results at different times. The disease identification results include one or more of the following: disease category, number of detected images, detection location, and identification confidence level. Based on the comparison results, the module determines the change type of the disease identification results within the building site unit. The change type includes one of the following: change from not detected to detected, change from detected to not detected, increase in the number of detected images, decrease in the number of detected images, and basically consistent detection status. A corresponding disease change record is then generated.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: First, this invention makes full use of crowdsourced images continuously accumulated by internet platforms and user uploads, enabling the acquisition of visual information related to architectural site defects without relying entirely on specialized field collection and professional surveying images, thereby improving the flexibility and sustainability of defect monitoring data acquisition.

[0017] Secondly, by using geographical reference and site unit affiliation, this invention associates crowdsourced images with scattered sources and inconsistent spatial orientations with specific architectural site units, thereby enhancing the spatial availability of crowdsourced images for architectural site monitoring.

[0018] Third, this invention uses the visual content of crowdsourced images to identify architectural defects, and can directly extract information such as defect type, approximate location and confidence level from the images, so that the defect identification results have clear image evidence to support them.

[0019] Fourth, this invention adopts a differentiated identification strategy for different disease morphologies. For crack-type diseases, a target detection method is used, and for irregularly shaped diseases such as collapse, peeling, and vegetation erosion, an image classification and sliding window combination method is used, which improves the adaptability of disease identification under complex crowdsourced image conditions.

[0020] Fifth, this invention conducts comparative analysis on crowdsourced images and disease identification results from different periods of the same architectural site unit, which can form a time-series record of disease identification results and support the expansion from single disease identification to continuous monitoring records.

[0021] Sixth, this invention constructs an image database of architectural ruins, which uniformly stores original images, ruin unit attribution results, ruin identification results, geographic reference information, time information, source information, and ruin change records, facilitating subsequent querying, statistics, mapping, and protection decisions. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart of a method for dynamic monitoring of architectural site defects based on crowdsourced imagery, provided in an embodiment of the present invention; Figure 2 A structural diagram of a dynamic monitoring device for architectural ruins based on crowdsourced imagery provided in an embodiment of the present invention; Figure 3 A flowchart for crowdsourced image georeferencing and site unit attribution provided in an embodiment of the present invention; Figure 4 This is a flowchart of disease identification and change detection provided in an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that: The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0026] like Figure 1 As shown in the figure, this specific embodiment provides a method for dynamic monitoring of architectural site defects based on crowdsourced images. It addresses the need for continuous monitoring of architectural site defects by acquiring relevant crowdsourced images of architectural sites from sources such as heritage-themed websites, open image platforms, street view or crowdsourced map platforms, outdoor trajectory platforms, and user uploads. By combining image metadata, architectural site vector data, and a defect identification model based on image visual content, it achieves continuous discovery, automatic identification, temporal change recording, and dynamic updating of the defect image library for architectural site defects. The method includes the following steps: Step S1, acquiring crowdsourced images and their metadata related to architectural sites at preset time intervals, and performing non-target scene filtering, duplicate image removal, and metadata standardization on candidate images to obtain a set of valid crowdsourced images; Step S2, based on the spatial association information carried by the valid crowdsourced images and combined with architectural site vector data, spatial reference and site unit affiliation are performed on the valid crowdsourced images to establish a correspondence between images and architectural site units; Step S3, based on the visual content of the valid crowdsourced images affixed to architectural site units, disease identification is performed, disease information in the architectural site is extracted, and disease category, disease location or response area, and identification confidence level are obtained; Step S4, based on the crowdsourced images and disease identification results of the same architectural site unit at different times, change detection is performed to determine the change type of the disease identification results and generate a disease change record; Step S5, the original images, disease identification results, geographic reference information, time information, source information, and disease change record are written into the architectural site disease image library, and the architectural site disease image library is dynamically updated according to the crowdsourced images acquired periodically thereafter. The specific implementation methods for each step are described in detail below.

[0027] Through the above implementation methods, crowdsourced images with scattered sources and inconsistent spatial orientations are transformed into structured monitoring data with spatial orientation, temporal attributes, and disease identification results. This effectively solves the technical problems of low efficiency of manual inspections, high cost of professional image acquisition, and difficulty in directly using crowdsourced images for continuous monitoring in the existing technology.

[0028] Step S1: Periodically acquire crowdsourced images and metadata related to architectural sites.

[0029] Specifically, step S1 includes the following steps.

[0030] Step S1.1: According to the preset collection cycle, based on the site unit number, spatial location, site name or route range in the architectural site vector data, obtain relevant images of the architectural site from heritage websites, open image platforms, street view or crowdsourced map platforms, outdoor trajectory platforms or user upload terminals.

[0031] For heritage-themed websites organized by site unit numbers or site entries, extract the site unit numbers or site name list from the architectural site vector data, and obtain the images under the corresponding site unit according to the site unit number or site name.

[0032] For open image platforms, street view or crowdsourced map platforms with geographic coordinates or supporting spatial retrieval, a retrieval range is constructed based on the spatial location of the site units in the architectural site vector data, and candidate images are obtained within the retrieval range.

[0033] For outdoor trajectory platforms organized by routes, tracks, or place names, candidate images are obtained based on the site name, place name keywords, route range, or trajectory clues. For user uploads, images of architectural sites uploaded by users, along with related information, are received.

[0034] Step S1.2: Extract and record the source information, time information, and spatial association information that can be used for spatial attribution of candidate images.

[0035] The source information includes the image source platform; the time information includes at least one of the acquisition time and upload time, used to determine the temporal position of the candidate image; the spatial association information includes at least one of the site unit number, geographic coordinates, route clues, and place name clues, used for spatial reference and attribution between subsequent images and architectural site units.

[0036] Step S1.3: Filter the candidate images for non-target scenes.

[0037] Due to the complexity of crowdsourced image sources, candidate images may include non-target scenes unrelated to the architectural site itself, such as roads, vehicles, tourists, modern buildings, and purely natural landscapes. Therefore, it is necessary to screen candidate images. The non-target scene filtering is mainly based on image content recognition and can be combined with manual review to remove images that do not contain the architectural site itself or are unrelated to the monitoring object of the architectural site.

[0038] Step S1.4: Perform duplicate image removal and metadata standardization on the candidate images.

[0039] For images that are repeatedly uploaded, reproduced, or highly similar across the same or different platforms, image similarity detection is used to remove duplicate or highly similar images. Subsequently, the source information, temporal information, and spatial association information of the candidate images are standardized and organized to obtain a valid set of crowdsourced images.

[0040] Step S2: Spatial reference and architectural heritage unit attribution are performed on the valid crowdsourced images.

[0041] Specifically, such as Figure 3 As shown, step S2 includes the following steps.

[0042] Step S2.1: For valid crowdsourced images with site unit numbers, read the site unit number and match it with the corresponding number field in the architectural site vector data to directly establish the association between the image and the corresponding architectural site unit.

[0043] Step S2.2: For valid crowdsourced images that do not have a site unit number but have geographic coordinates, read their latitude and longitude coordinates, calculate the spatial distance between the image coordinates and each site unit in the building site vector data, and assign the image to the building site unit that is closest to it and meets the preset distance threshold.

[0044] Step S2.3: For valid crowdsourced images that do not have site unit numbers and clear geographical coordinates but have route clues or place name clues, the corresponding architectural site unit is determined by combining the route range, place name pointing information and architectural site vector data for auxiliary interpretation.

[0045] In this step, the architectural site vector data is point, line, or area vector data, and its attribute fields include information such as site unit number, site name, site type, and spatial location. This architectural site vector data serves as a baseline data for effective crowdsourced image spatial reference and site unit attribution.

[0046] After the site unit is assigned, one or more of the following information is associated and stored: image number, site unit number, assignment method, coordinate information, matching distance, and matching confidence level, to form a spatial reference relationship between the image and the architectural site unit.

[0047] Step S3: Identify architectural defects in the assigned crowdsourced images.

[0048] Specifically, such as Figure 4 As shown, step S3 includes the following steps.

[0049] Step S3.1: Construct a building site disease identification model based on image visual content.

[0050] In this specific embodiment, the disease identification model includes a target detection model and an image classification model. The types of architectural site diseases include at least one of cracks, collapses, peeling, and vegetation erosion. The disease identification model takes the visual content of the architectural site image as input to extract the disease category, disease location or response area, and identification confidence level.

[0051] The disease identification model is obtained through pre-training. The training samples are derived from images of architectural ruins that have been assigned to ruin units or confirmed manually, and are labeled according to the type of disease. Specifically, for crack-type diseases, labeled samples containing information on the location, extent, and category of cracks are used to train the target detection model; for collapse, peeling, and vegetation erosion-type diseases, classification samples are constructed based on disease areas or local image patches to train the image classification model.

[0052] Step S3.2: For crack-type defects with relatively clear boundaries, a target detection model is used to identify the effective crowdsourced images and output the location range, category, and confidence level of the crack defects.

[0053] Step S3.3: For diseases with irregular shapes or whose boundaries are difficult to define precisely, such as collapse, peeling, and vegetation erosion, an image classification model combined with a sliding window method is used to identify local diseases.

[0054] Specifically, the input image is scanned locally according to a preset window size and step size to obtain multiple local image regions; the local image regions are input into an image classification model to determine the disease category, and suspected disease responses are screened based on classification confidence.

[0055] Step S3.4: Confidence screening, duplicate response merging, and result integration are performed on the disease identification results in the same image to obtain the disease identification results in a single image.

[0056] Step S3.5: After completing the disease identification of a single image, associate the disease category, disease location or response area, identification confidence level and image number with the corresponding architectural site unit to form architectural site unit-level disease identification results.

[0057] Step S4: Detect changes based on the disease identification results at different times.

[0058] Specifically, step S4 includes the following steps.

[0059] Step S4.1: Organize the crowdsourced images and their disease identification results acquired at different times according to the building site unit number and image time information.

[0060] Step S4.2: For the same architectural site unit, compare the disease identification results from different periods.

[0061] The disease identification results include disease category, number of detected images, detection location, and identification confidence level.

[0062] Step S4.3: Determine the type of change in the identification results of defects within the architectural site unit based on the comparison results.

[0063] The types of change include changes from undetected to detected, changes from detected to undetected, an increase in the number of detected cases, a decrease in the number of detected cases, and a situation where the detection status remains basically the same. These types of change are used to describe the temporal changes in disease identification results at different times and are not directly equivalent to the precise physical changes in the disease entity.

[0064] Step S4.4: After determining the type of change, organize the building site unit number, image time information, disease category, disease identification result, change type and identification confidence into a disease change record, and associate the disease change record with the corresponding building site unit.

[0065] Step S5: Construct and dynamically update the image database of architectural ruins damage.

[0066] Specifically, step S5 includes the following steps.

[0067] Step S5.1: Establish an image database of architectural ruins.

[0068] The architectural site disease image database is used to uniformly store and manage original crowdsourced images, architectural site unit information, the relationship between images and site units, disease identification results, disease change records, and data source records.

[0069] Step S5.2: Write the original image records, site unit records, image-site unit association records, disease identification result records, disease change records, and data source records into the architectural site disease image database.

[0070] Among them, the original image record is used to save valid crowdsourced images and their basic metadata; the site unit record is used to save the building site unit number, name and spatial location; the image and site unit association record is used to save the image attribution method and matching distance; the disease identification result record is used to save the disease category, disease location or response area and identification confidence level; the disease change record is used to save the change type of disease identification results at different times; and the data source record is used to save the image source platform, acquisition time or upload time.

[0071] Step S5.3: When the system acquires new crowdsourced images in subsequent monitoring cycles, repeat steps S1 to S4, and add the new image attribution results, disease identification results, and disease change records to the building site disease image database to realize the dynamic updating of the building site disease image database.

[0072] like Figure 2 As shown in the illustration, this specific embodiment also provides a dynamic monitoring device for architectural ruins based on crowdsourced imagery. The device includes an image acquisition module, a geographic reference module, a ruin identification module, a change detection module, a ruin image database construction module, and a result output module.

[0073] The image acquisition module is used to acquire crowdsourced images and their metadata related to architectural sites at preset time intervals, and to perform non-target scene filtering, duplicate image removal and metadata standardization on candidate images to obtain an effective set of crowdsourced images.

[0074] The georeference module is used to georeference and assign site units to valid crowdsourced images based on the spatial association information carried by the images and combined with the vector data of architectural sites, and to establish the correspondence between the images and architectural site units.

[0075] The disease identification module is used to extract disease information of architectural ruins based on the visual content of effective crowdsourced images, and obtain the disease category, disease location or response area and identification confidence level.

[0076] The change detection module is used to compare the disease identification results of the same architectural site unit at different times and generate a record of disease changes.

[0077] The disease image database construction module is used to uniformly store original images, site unit information, the relationship between images and site units, disease identification results, time information, source information, and disease change records, and to dynamically update them based on subsequently acquired crowdsourced images.

[0078] The results output module is used to output the monitoring results of architectural ruins, records of changes in the results of disease identification, statistical results of diseases at the ruins unit level, or results of the spatial distribution of diseases.

[0079] The specific implementation of each module in the device described in this embodiment can be achieved by referring to the method steps, and will not be elaborated here.

[0080] In summary, this invention discloses a method and apparatus for dynamic monitoring of architectural site defects based on crowdsourced imagery. The method acquires relevant crowdsourced images and their metadata for architectural sites at preset time intervals, and performs non-target scene filtering, duplicate image removal, and metadata standardization on candidate images to obtain a valid set of crowdsourced images. Based on the spatial association information carried by the valid crowdsourced images, spatial reference and site unit attribution are performed in conjunction with architectural site vector data. Defect identification is performed based on the visual content of the assigned images, extracting defect categories, defect locations or response areas, and identification confidence levels. Defect identification results for the same architectural site unit at different times are compared to generate a defect change record. The relevant data is then written into an architectural site defect image database for dynamic updates. This invention can improve the efficiency of architectural site defect information acquisition, spatial organization, and continuous monitoring.

[0081] This invention can transform scattered crowdsourced images into monitoring data for architectural ruins with spatial orientation, temporal attributes, and disease identification results, enabling continuous discovery, automatic identification, temporal change recording, and dynamic updating of the disease image database for architectural ruins. It is applicable to dynamic monitoring of diseases of architectural ruins such as the Great Wall, city walls, ancient architectural ruins, stone ruins, and earthen ruins.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic monitoring of architectural site defects based on crowdsourced imagery, characterized in that, include: Step S1: Acquire crowdsourced images and their metadata related to the architectural site at preset time intervals, and perform non-target scene filtering, duplicate image removal and metadata standardization on the candidate images to obtain an effective crowdsourced image set. Step S2: Based on the spatial association information carried by the effective crowdsourced images and combined with the architectural site vector data, perform spatial reference and site unit attribution on the effective crowdsourced images, and establish the correspondence between the images and architectural site units. Step S3: Based on the visual content of the valid crowdsourced images belonging to the architectural heritage unit, identify the defects, extract the defect information in the architectural heritage, and obtain the defect category, defect location or response area and identification confidence level. Step S4: Based on the crowdsourced images and their disease identification results of the same architectural site unit at different times, perform change detection, determine the change type of the disease identification results, and generate a disease change record; Step S5: Write the original image, disease identification results, geographic reference information, time information, source information, and disease change records into the architectural ruins disease image database, and dynamically update the architectural ruins disease image database based on the crowdsourced images obtained periodically.

2. The method for dynamic monitoring of architectural site defects based on crowdsourced imagery according to claim 1, characterized in that, Step S2 includes: For valid crowdsourced images with site unit numbers, read the site unit number and match it with the corresponding number field in the architectural site vector data to directly establish the association between the image and the corresponding architectural site unit. For valid crowdsourced images that do not have site unit numbers but have geographic coordinates, calculate the spatial distance between the image coordinates and each site unit in the architectural site vector data, and assign the image to the architectural site unit that is closest to the preset distance threshold. For valid crowdsourced images that do not have site unit numbers and clear geographical coordinates but have route clues or place name clues, the corresponding architectural site unit is determined by combining the route range, place name information, and architectural site vector data.

3. The method for dynamic monitoring of architectural site defects based on crowdsourced imagery according to claim 1, characterized in that, Step S3 includes: A model for identifying architectural defects based on visual image content is constructed. The defect identification model includes a target detection model and an image classification model. A target detection model is used to identify crack-type defects in valid crowdsourced images, and the location range, category and confidence level of the crack defects are output. An image classification model combined with a sliding window approach was used to identify local diseases such as collapse, peeling, and vegetation erosion in valid crowdsourced images. The disease identification results in the same image are subjected to confidence screening, duplicate response merging and result integration to obtain the disease identification results in a single image. By associating the disease category, disease location or response area, identification confidence level and image number in a single image with the corresponding architectural site unit, a disease identification result at the architectural site unit level is formed.

4. The method for dynamic monitoring of architectural site defects based on crowdsourced imagery according to claim 1, characterized in that, Step S4 includes: Based on the architectural site unit number and image time information, the crowdsourced images and their disease identification results acquired at different times are organized chronologically. For the same architectural site unit, the disease identification results from different periods are compared. The disease identification results include one or more of the following: disease category, number of detected images, detection location, and identification confidence level. Based on the comparison results, determine the type of change in the identification results of defects within the architectural site unit. The type of change includes one of the following: change from not detected to detected, change from detected to not detected, increase in the number of detected defects, decrease in the number of detected defects, and basically consistent detection status. Generate a corresponding defect change record.

5. The method for dynamic monitoring of architectural site defects based on crowdsourced imagery according to claim 1, characterized in that, Step S5 includes: Establish an image database of architectural ruins, which includes original image records, ruin unit records, image-rule association records, ruin identification result records, ruin change records, and data source records. The original image records, site unit records, image-site unit association records, disease identification result records, disease change records, and data source records are written into the architectural site disease image database. Once new crowdsourced images are acquired, steps S1 to S4 are repeated, and the new original images, image-to-site unit association records, disease identification results records, and disease change records are added to the architectural site disease image database to achieve dynamic updates of the architectural site disease image database.

6. The method for dynamic monitoring of architectural site defects based on crowdsourced imagery according to claim 1, characterized in that, Step S1 includes: According to the preset collection cycle, based on the site unit number, spatial location, site name or route range in the vector data of the architectural site, relevant images of the architectural site are obtained from heritage websites, open image platforms, street view or crowdsourced map platforms, outdoor trajectory platforms or user upload terminals. Extract and record the source information, time information, and spatial association information for spatial attribution of candidate images; wherein, the source information includes the image source platform, the time information includes at least one of the acquisition time and upload time, and the spatial association information includes at least one of the site unit number, geographic coordinates, route clues, and place name clues; Candidate images are filtered for non-target scenes to remove images that do not contain the building site itself or are unrelated to the building site monitoring object. The candidate images are subjected to duplicate image removal and metadata standardization processing to remove duplicate or highly similar images. The source information, time information and spatial correlation information of the candidate images are standardized and organized to obtain an effective crowdsourced image set.

7. A dynamic monitoring device for architectural ruins based on crowdsourced imagery, characterized in that, include: The image acquisition module is used to acquire crowdsourced images and their metadata related to architectural sites at preset time intervals, perform non-target scene filtering, duplicate image removal and metadata standardization on candidate images, and obtain an effective set of crowdsourced images. The georeference module is used to perform spatial reference and site unit assignment on valid crowdsourced images based on the spatial association information carried by the images and combined with the vector data of architectural sites, and to establish the correspondence between the images and architectural site units. The disease identification module is used to identify diseases based on the visual content of valid crowdsourced images belonging to the building site unit, extract disease information in the building site, and output the disease category, disease location or response area and identification confidence level. The change detection module is used to detect changes based on crowdsourced images and their disease identification results of the same architectural site unit at different times, determine the type of change in the disease identification results, and generate a disease change record. The disease image database construction module is used to write the original images, disease identification results, geographic reference information, time information, source information and disease change records into the architectural site disease image database, and dynamically update the architectural site disease image database based on the crowdsourced images acquired periodically.

8. The dynamic monitoring device for architectural ruins based on crowdsourced imagery according to claim 7, characterized in that, For valid crowdsourced images with site unit numbers, the geographic reference module reads the site unit number and matches it with the corresponding number field in the architectural site vector data to directly establish the association between the image and the corresponding architectural site unit. For valid crowdsourced images without site unit numbers but with geographic coordinates, the module calculates the spatial distance between the image coordinates and each site unit in the architectural site vector data, and assigns the image to the nearest architectural site unit that meets a preset distance threshold. For valid crowdsourced images without site unit numbers and explicit geographic coordinates but with route clues or place name clues, the module uses the route range, place name information, and architectural site vector data for auxiliary interpretation to determine the corresponding architectural site unit.

9. The dynamic monitoring device for architectural ruins based on crowdsourced imagery according to claim 7, characterized in that, The disease identification module includes a target detection model and an image classification model. The target detection model is used to identify crack-type diseases in valid crowdsourced images and output the location range, category, and confidence level of the crack-type diseases. The image classification model is used to identify local diseases such as collapse, peeling, and vegetation erosion in valid crowdsourced images using a sliding window method. The disease identification module is also used to perform confidence filtering, duplicate response merging, and result integration on the disease identification results in the same image to obtain the disease identification results in a single image. The disease category, disease location or response area, identification confidence level, and image number in a single image are associated with the corresponding architectural heritage unit to form architectural heritage unit-level disease identification results.

10. The dynamic monitoring device for architectural ruins based on crowdsourced imagery according to claim 7, characterized in that, The change detection module organizes crowdsourced images and their disease identification results acquired at different times according to the building site unit number and image time information. For the same building site unit, it compares the disease identification results at different times. The disease identification results include one or more of the following: disease category, number of detected images, detection location, and identification confidence level. Based on the comparison results, it determines the change type of the disease identification results within the building site unit. The change type includes one of the following: change from not detected to detected, change from detected to not detected, increase in the number of detected images, decrease in the number of detected images, and basically consistent detection status. The module then generates a corresponding disease change record.