An engineering management method based on automatic analysis of construction site watermark image information

Abstract

The application discloses an engineering management method based on automatic analysis of construction site watermark image information, imports a construction site watermark photo file to be processed, carries out image preprocessing on the watermark photo file to obtain standardized images and boundary coordinates of a watermark text area, carries out area cutting on the images according to the boundary coordinates of the watermark text area, carries out OCR recognition on image subblocks obtained through the cutting, generates corresponding structured data, automatically renames and classifies archives of the watermark photo according to the structured data, and automatically generates a construction log PDF file according to the structured data and a preset document template. The application has the beneficial effects that the resolution standardization and image enhancement processing improve the accuracy of watermark text recognition, improve the stability of image processing and realize batch image processing, the automatic naming and classified archiving of the photo are realized, manual intervention is reduced, the construction log PDF file is automatically generated, and the automation, standardization and traceability of construction site information management are improved.

Description

Technical Field

[0001] This invention relates to the fields of digital image processing technology, optical character recognition technology, and engineering information management technology, and in particular to an engineering management method based on automatic parsing of watermark image information at construction sites. Background Technology

[0002] In the construction management process, the construction site is typically documented by taking photos with watermarked information. This watermarked information generally includes key management data such as the construction area, construction content, shooting time, and labor information.

[0003] In current technologies, the management of construction site photos mainly relies on manual viewing, recording, and archiving. This is not only inefficient but also prone to problems such as information omissions, non-standard file naming, and disorganized archiving. When the project is large in scale and has a large number of photos, traditional manual management methods are difficult to meet the needs of efficient, standardized, and traceable project management.

[0004] In addition, the shooting environment at the construction site is complex, with large variations in lighting conditions and inconsistent photo resolution. Although the position of the watermark area is relatively fixed, there are still some deviations, resulting in low recognition accuracy when directly performing OCR recognition and making it difficult to achieve stable automated processing.

[0005] Therefore, it is necessary to propose an engineering management method that can standardize the processing of watermarked photos at construction sites, automatically identify them, and generate structured management data to address the aforementioned technical issues, thereby improving the automation and accuracy of construction information management. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, the present invention aims to provide an engineering management method based on the automatic parsing of watermarked image information at construction sites. By performing standardized preprocessing, regional parameterized positioning, OCR recognition, and structured data construction on watermarked photos at construction sites, the method enables automatic naming, automatic archiving, and automatic generation of construction logs, thereby improving the efficiency and standardization of engineering management.

[0007] An engineering management method based on automatic parsing of watermarked image information at construction sites includes the following steps: S1. Import the watermarked photo file of the construction site to be processed, and perform image preprocessing on the watermarked photo file to obtain the boundary coordinates of the standardized image and watermark text area. S2. Based on the boundary coordinates of the watermarked text region, the image is cropped, and the cropped image sub-blocks are subjected to OCR recognition to generate corresponding structured data. S3. Automatically rename and hierarchically archive the watermarked photos based on the structured data, and generate construction information ledger data; S4. Automatically generate a construction log PDF file by calling a preset document template based on the structured data.

[0008] Step S1 includes: S11. Standardize the resolution of watermarked photos at the construction site to unify the images to the preset standard resolution. S12. Perform image enhancement processing on the resolution-normalized image to improve the contrast and clarity of the text area; S13. Locate the watermark text region based on manually marked regional coordinate parameters and output the boundary coordinates of the watermark text region.

[0009] Step S11 includes: S111, Obtain the width of the original image. ,high And pixel density information, set the preset standard resolution pixel size as ; S112. Calculate the scaling ratio and width scaling ratio based on the preset standard resolution. Height scaling ratio ; S113. The image is scaled using a bicubic interpolation algorithm. The weight function of the bicubic interpolation algorithm is: ; The formula for two-dimensional interpolation merging is: , in The pixel coordinates to be interpolated are... For decimal offset, The original pixel values ​​are the 4×4 neighborhood around the point to be interpolated. The algorithm calculation steps are as follows: First, select 4×4 grid points around the point to be interpolated, substitute them into the weight function to calculate the interpolation weights in the x and y directions respectively, perform point-by-point interpolation in the first direction, and then perform column direction interpolation to obtain the scaled target pixel values. S114. Sharpen the scaled image using a sharpening filter to reduce the loss of sharpness caused by the resolution change.

[0010] Step S12 includes: S121. Convert the image to a grayscale image. The grayscale conversion formula is: ,in These are the red, green, and blue channel values ​​of the original image pixels, respectively; The target grayscale image pixel value; S122. Binarization is performed using an adaptive threshold segmentation method. The local threshold calculation formula for this algorithm is as follows: ,in In pixels Centered The average gray level within the neighborhood; The threshold adjustment coefficient is used; the binarization determination calculation formula is as follows: ; in The calculation steps are as follows: traverse all pixels of the grayscale image, calculate the grayscale mean value in the neighborhood of its preset size for each pixel, substitute it into the local threshold formula to obtain the adaptive threshold of the pixel, and then generate the binary image pixel value point by point according to the binarization determination rules. S123. Perform contrast enhancement and noise suppression processing on the binarized image to improve the accuracy of subsequent text recognition.

[0011] Step S13 includes: S131. During the system initialization phase, sample watermark photos are marked with regions through manual interaction, and the boundary coordinates of the watermark text region are recorded. ; S132. Save the boundary coordinates as clipping region parameters in the configuration file; S133. During the batch processing stage, the cropping region parameters are read, and the region coordinates are proportionally adapted according to the current image size. The adaptation calculation formula is as follows: ,in , which is the scaling ratio of the width and height of the current image compared to the sample image; To adapt the target boundary coordinates; S134. Based on the adapted region coordinates, crop and locate the image, and output the boundary coordinates of the watermark text region.

[0012] Step S2 includes: S21. Based on the boundary coordinates of the watermark text region output by S1, crop the image to obtain image sub-blocks with multiple fixed functional regions. S22. The image sub-blocks are input into an OCR text recognition model for recognition. The OCR text recognition model is composed of the DBNet text detection model and the CRNN text recognition model in the PaddleOCR framework; wherein the differentiable binarization calculation formula of DBNet is: , This is the magnification factor; These are the pixel values ​​of the probability map. The threshold image pixel values; the CTC loss calculation formula for CRNN is: ; in To map to the target sequence The set of all paths The path probability is calculated using the PaddleOCR model algorithm. First, text regions are detected in image sub-blocks using the DBNet text detection network. After feature fusion and bi-branch prediction, a text region probability map and a threshold map are obtained. These are then substituted into the differentiable binarization formula to generate an approximate binary map. After contour extraction and NMS non-maximum suppression, an accurate text detection box is obtained. The text image within the detection box is then input into a CRNN. ​​After CNN convolutional feature extraction, Bi-LSTM bidirectional recurrent sequence modeling, and Softmax classification, the sequence alignment and character recognition are completed by substituting into the CTC loss formula, and the text character sequence is output. S23. Extract text and filter null values ​​from the OCR recognition results, and concatenate multiple lines of text within the same area. S24. According to the preset area function mapping rules, the recognition results are mapped to the construction area field, construction content field, shooting time field and labor information field respectively. S25. Clean up and standardize the text of each field to create a structured data object.

[0013] Step S3 includes: S31. Based on the construction area, construction content, and shooting time fields in the structured data, automatically rename the watermarked photos according to the naming rule of "construction area-construction content-shooting date". S32. When a newly generated filename contains duplicates, an incrementing suffix is ​​added to avoid duplicate names; the formula for duplicate name correction is as follows: Where N is the original renaming result, n is a positive integer incrementing, n=1,2,3..., and N′ is the final unique filename. S33. Based on the construction area and shooting time fields, automatically create multi-level folders according to the hierarchical directory structure of "construction area / year / month"; S34. Archive the renamed watermarked photos to the corresponding folder; S35. Generate a construction ledger data table containing fields for construction area, construction content, shooting time, and number of laborers.

[0014] Step S4 includes: S41. Generate construction log text data based on structured data; convert the structured data into log text content according to preset field mapping rules, with the mapping operation relationship as follows: ,in These fields are: construction area, construction content, shooting time, and labor information. This is a predefined mapping function from fields to log text. S42. Fill the construction log text data into the preset document template variable field; S43. Use the LaTeX compilation engine to generate PDF format construction log files; S44. Name the generated construction log PDF files according to the construction time and store them in the corresponding construction log folder.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the engineering management method based on automatic parsing of watermark image information at the construction site.

[0016] Compared with existing technologies, the present invention has the following advantages: by standardizing resolution and enhancing image processing, the accuracy of watermark text recognition is improved; by adopting a manual initialization marking and parameterized region reuse mechanism, the stability of image processing is improved and batch image processing is achieved; automatic naming and hierarchical archiving of photos are realized, reducing manual intervention; construction log PDF files are automatically generated, realizing a closed loop of construction management data; and the automation, standardization and traceability of construction site information management are improved. Attached Figure Description

[0017] Figure 1 This is a flowchart of the rapid construction method for pile foundations according to the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0019] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0020] like Figure 1 As shown, an engineering management method based on automatic parsing of watermark image information at construction sites includes the following steps: S1. Import the watermarked photo file of the construction site to be processed, and perform image preprocessing on the watermarked photo file to obtain the boundary coordinates of the standardized image and watermark text area. S2. Based on the boundary coordinates of the watermarked text region, the image is cropped, and the cropped image sub-blocks are subjected to OCR recognition to generate corresponding structured data. S3. Automatically rename and hierarchically archive the watermarked photos based on the structured data, and generate construction information ledger data; S4. Automatically generate a construction log PDF file by calling a preset document template based on the structured data.

[0021] Step S1 includes: S11. Standardize the resolution of watermarked photos at the construction site to unify the images to the preset standard resolution. S12. Perform image enhancement processing on the resolution-normalized image to improve the contrast and clarity of the text area; S13. Locate the watermark text region based on manually marked regional coordinate parameters and output the boundary coordinates of the watermark text region.

[0022] Step S11 includes: S111, Obtain the width of the original image. ,high And pixel density information, set the preset standard resolution pixel size as ; S112. Calculate the scaling ratio and width scaling ratio based on the preset standard resolution. Height scaling ratio ; S113. The image is scaled using a bicubic interpolation algorithm. The weight function of the bicubic interpolation algorithm is: ; The formula for two-dimensional interpolation merging is: , in The pixel coordinates to be interpolated are... For decimal offset, The original pixel values ​​are the 4×4 neighborhood around the point to be interpolated. The algorithm calculation steps are as follows: First, select 4×4 grid points around the point to be interpolated, substitute them into the weight function to calculate the interpolation weights in the x and y directions respectively, perform point-by-point interpolation in the first direction, and then perform column direction interpolation to obtain the scaled target pixel values. S114. Sharpen the scaled image using a sharpening filter to reduce the loss of sharpness caused by the resolution change.

[0023] Step S12 includes: S121. Convert the image to a grayscale image. The grayscale conversion formula is: ,in These are the red, green, and blue channel values ​​of the original image pixels, respectively; The target grayscale image pixel value; S122. Binarization is performed using an adaptive threshold segmentation method. The local threshold calculation formula for this algorithm is as follows: ,in In pixels Centered The average gray level within the neighborhood; The threshold adjustment coefficient is used; the binarization determination calculation formula is as follows: ; in The calculation steps are as follows: traverse all pixels of the grayscale image, calculate the grayscale mean value in the neighborhood of its preset size for each pixel, substitute it into the local threshold formula to obtain the adaptive threshold of the pixel, and then generate the binary image pixel value point by point according to the binarization determination rules. S123. Perform contrast enhancement and noise suppression processing on the binarized image to improve the accuracy of subsequent text recognition.

[0024] Step S13 includes: S131. During the system initialization phase, sample watermark photos are marked with regions through manual interaction, and the boundary coordinates of the watermark text region are recorded. ; S132. Save the boundary coordinates as clipping region parameters in the configuration file; S133. During the batch processing stage, the cropping region parameters are read, and the region coordinates are proportionally adapted according to the current image size. The adaptation calculation formula is as follows: ,in , which is the scaling ratio of the width and height of the current image compared to the sample image; To adapt the target boundary coordinates; S134. Based on the adapted region coordinates, crop and locate the image, and output the boundary coordinates of the watermark text region.

[0025] Step S2 includes: S21. Based on the boundary coordinates of the watermark text region output by S1, crop the image to obtain image sub-blocks with multiple fixed functional regions. S22. The image sub-blocks are input into an OCR text recognition model for recognition. The OCR text recognition model is composed of the DBNet text detection model and the CRNN text recognition model in the PaddleOCR framework; wherein the differentiable binarization calculation formula of DBNet is: , This is the magnification factor; These are the pixel values ​​of the probability map. The threshold image pixel values; the CTC loss calculation formula for CRNN is: ; in To map to the target sequence The set of all paths The path probability is calculated using the PaddleOCR model algorithm. First, text regions are detected in image sub-blocks using the DBNet text detection network. After feature fusion and bi-branch prediction, a text region probability map and a threshold map are obtained. These are then substituted into the differentiable binarization formula to generate an approximate binary map. After contour extraction and NMS non-maximum suppression, an accurate text detection box is obtained. The text image within the detection box is then input into a CRNN. ​​After CNN convolutional feature extraction, Bi-LSTM bidirectional recurrent sequence modeling, and Softmax classification, the sequence alignment and character recognition are completed by substituting into the CTC loss formula, and the text character sequence is output. S23. Extract text and filter null values ​​from the OCR recognition results, and concatenate multiple lines of text within the same area. S24. According to the preset area function mapping rules, the recognition results are mapped to the construction area field, construction content field, shooting time field and labor information field respectively. S25. Clean up and standardize the text of each field to create a structured data object.

[0026] Step S3 includes: S31. Based on the construction area, construction content, and shooting time fields in the structured data, automatically rename the watermarked photos according to the naming rule of "construction area-construction content-shooting date". S32. When a newly generated filename contains duplicates, an incrementing suffix is ​​added to avoid duplicate names; the formula for duplicate name correction is as follows: Where N is the original renaming result, n is a positive integer incrementing, n=1,2,3..., and N′ is the final unique filename. S33. Based on the construction area and shooting time fields, automatically create multi-level folders according to the hierarchical directory structure of "construction area / year / month"; S34. Archive the renamed watermarked photos to the corresponding folder; S35. Generate a construction ledger data table containing fields for construction area, construction content, shooting time, and number of laborers.

[0027] Step S4 includes: S41. Generate construction log text data based on structured data; convert the structured data into log text content according to preset field mapping rules, with the mapping operation relationship as follows: ,in These fields are: construction area, construction content, shooting time, and labor information. This is a predefined mapping function from fields to log text. S42. Fill the construction log text data into the preset document template variable field; S43. Use the LaTeX compilation engine to generate PDF format construction log files; S44. Name the generated construction log PDF files according to the construction time and store them in the corresponding construction log folder.

[0028] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the engineering management method based on automatic parsing of watermark image information at the construction site.

[0029] Compared with existing technologies, the present invention has the following advantages: by standardizing resolution and enhancing image processing, the accuracy of watermark text recognition is improved; by adopting a manual initialization marking and parameterized region reuse mechanism, the stability of image processing is improved and batch image processing is achieved; automatic naming and hierarchical archiving of photos are realized, reducing manual intervention; construction log PDF files are automatically generated, realizing a closed loop of construction management data; and the automation, standardization and traceability of construction site information management are improved.

[0030] Example 1: Application on the construction site of the main structure of a residential building This embodiment is applied to the construction site of a high-rise residential building. The project has 18 floors and is currently in the construction phase of the main structure from the 5th to the 8th floor. No less than 200 watermarked photos are taken at the construction site every day. The watermark information includes the construction area (such as "5th floor east unit" and "6th floor west unit"), the construction content (such as "formwork installation", "rebar tying" and "concrete pouring"), the shooting time (accurate to the minute), and the labor information (such as "8 steelworkers and 6 carpenters"). The watermark area is fixed in the lower right corner of the photo and is about 1 / 4 of the image width and 1 / 5 of the height. The shooting environment has problems such as uneven lighting (backlight in the morning and evening, strong light at noon) and different photo resolutions (1080P taken with a mobile phone and 4K taken with a camera). The method of this invention is used to realize automatic photo management and construction log generation.

[0031] Specific implementation steps S1: Image preprocessing (obtaining standardized image and watermark boundary coordinates) S11: Resolution Normalization Processing. Obtain the original image resolution. For mobile phone images, the resolution is 1920×1080; for camera images, the resolution is 3840×2160. Set the preset standard resolution to 1920×1080 (balancing recognition efficiency and clarity). Calculate the scaling ratio: For 4K images, the width scaling ratio = 1920 / 3840 = 0.5, and the height scaling ratio = 1080 / 2160 = 0.5. Use bicubic interpolation for scaling. After scaling, perform sharpening filtering (using the Laplacian sharpening operator) to reduce sharpness loss.

[0032] S12: Image enhancement processing. The standardized image is converted into a grayscale image, which represents the pixel values ​​of the red, green, and blue channels of the original image; adaptive threshold segmentation is used for binarization, with a neighborhood size of 15×15. Histogram equalization is applied to the binarized image to enhance contrast, and median filtering (window size 3×3) is used to suppress noise and improve the clarity of the watermark text.

[0033] S13: Watermark Region Location. During system initialization, the watermark region boundary coordinates of three sample watermark photos are manually marked. The sample image resolution is 1920×1080, and the marked boundary coordinates are (1440, 864, 1920, 1080) (top left x, y, bottom right x, y). These coordinates are saved as cropping region parameters in the configuration file. During batch processing, these parameters are read, and proportional adaptation is performed according to the current image size (in this embodiment, some images have been standardized to 1920×1080, with an adaptation ratio of 1:1, and the marked coordinates are directly used; unstandardized images are adapted according to the scaling ratio), and the watermark text region boundary coordinates are output.

[0034] S2: OCR Recognition and Structured Data Generation S21: Region cropping. The standardized image is cropped based on the boundary coordinates (1440, 864, 1920, 1080) output by S1 to obtain four image sub-blocks with fixed functional regions, corresponding to the construction area, construction content, shooting time, and labor information, respectively (each sub-block is approximately 120×54 in size).

[0035] S22: OCR Recognition. Input four image sub-blocks into the PaddleOCR model; the model first extracts text features through DBnet, generates an approximate binary image and obtains accurate detection boxes, and then completes character recognition through CRNN, outputting a text sequence. For example, the recognition result of a photo is: construction area "5th floor east unit", construction content "rebar tying", shooting time "2026-02-20 09:35", and labor information "8 rebar workers and 6 carpenters".

[0036] S23-S25: ​​Text Processing and Structured Data Construction. The OCR recognition results are filtered for null values ​​(empty characters from fuzzy recognition), and multiple lines of text within the same area are concatenated (in this embodiment, the watermark text is single-line and does not require concatenation); according to preset area function mapping rules, the recognition results are mapped to corresponding fields; illegal characters are cleaned from the fields (irrelevant symbols such as "、", "," are removed, and the time format is unified as "YYYY-MM-DD HH:MM"), and a structured data object is constructed: {"Construction Area":"5th Floor East Unit","Construction Content":"Rebar Binding","Shooting Time":"2026-02-20 09:35","Labor Information":"8 Rebar Workers, 6 Carpenters"}.

[0037] S3: Automatic photo renaming, archiving, and ledger generation S31-S32: Automatic renaming. Following the naming convention of "Construction Area - Construction Content - Shooting Date", generate filenames such as "5th Floor East Unit - Rebar Binding - 20260220". If duplicate names exist (multiple photos of the same content from the same area on the same date), append suffixes using an incrementing count method, for example, "5th Floor East Unit - Rebar Binding - 20260220_1" and "5th Floor East Unit - Rebar Binding - 20260220_2".

[0038] S33-S34: Hierarchical archiving. Following a hierarchical directory structure of "Construction Area / Year / Month," create multi-level folders such as "5th Floor East Unit / 2026 / 02" and "6th Floor West Unit / 2026 / 02," and archive renamed photos to the corresponding folders for categorized management.

[0039] S35: Ledger Generation. Generate a construction information ledger data table, including fields: construction area, construction content, shooting time, and number of workers (extracting the number of people in the labor information, such as "14 people"), and update it to the project management system in real time for easy query and statistics by managers.

[0040] S4: Automatic generation of construction log PDF S41: Log Text Generation. According to the preset field mapping rules, convert structured data into log text. The generated text reads: "Today (2026-02-20 09:35), rebar tying work was carried out in the east unit area on the 5th floor. A total of 14 people participated in the work. The construction process was smooth, and there were no safety hazards on site."

[0041] S42-S44: PDF Generation and Storage. Fill the log text into the preset construction log template (including modules such as project name, construction date, construction area, construction content, and labor information), call the LaTeX compilation engine to generate a PDF file, name it "2026-02-20-5th Floor East Unit - Rebar Binding", and save it to the "Construction Log / 2026 / 02" folder.

[0042] In this embodiment, the processing time for 200 watermarked photos per day is reduced from 4 hours of manual management to 30 minutes, the OCR recognition accuracy rate reaches 98.5% (solving the recognition error caused by uneven lighting and different resolutions), the photo naming and archiving standardization rate is 100%, the efficiency of construction log generation is improved by 80%, effectively reducing manual intervention, realizing the automated and standardized management of construction information, and facilitating the traceability of the project in the later stage.

[0043] Example 2: Application in municipal roadbed construction site This embodiment is applied to the construction site of a municipal road expansion project. The road is 2.5km long and is currently in the roadbed backfilling and compaction stage. More than 150 watermarked photos are taken daily at the construction site. The watermark information includes the construction area (e.g., "K0+200-K0+300 section", "K0+500-K0+600 section"), construction content (e.g., "roadbed backfilling", "compaction testing", "drainage pipe pre-laying"), shooting time, and labor information (e.g., "12 general workers, 2 road roller drivers"). The watermark area is fixed in the lower left corner of the photo, with a size of approximately 1 / 3 of the image width and 1 / 6 of the height. The shooting environment has problems such as dust interference and insufficient lighting on cloudy days. The photo resolution is uniformly 2560×1440. The method of this invention achieves efficient management.

[0044] Specific implementation steps S1: Image preprocessing (obtaining standardized image and watermark boundary coordinates) S11: Resolution Normalization Processing. The original image resolution is 2560×1440. The preset standard resolution is set to 2560×1440 (no scaling required). The original image is then directly subjected to sharpening filtering (using the Gaussian sharpening operator) to improve text clarity.

[0045] S12: Image enhancement processing; Adaptive histogram equalization is used to enhance contrast, and Gaussian filtering (standard deviation = 1.5) is used to suppress noise caused by dust and improve text blurring caused by insufficient lighting on cloudy days.

[0046] S13: Watermark Region Location. During system initialization, the watermark region boundary coordinates (100, 1120, 850, 1440) of two sample photos are manually marked and saved as configuration file parameters. During batch processing, since the original image resolution is consistent with the sample photos, these coordinates are directly used to output the watermark text region boundary coordinates.

[0047] S2: OCR Recognition and Structured Data Generation S21: Region cropping. The image is cropped according to the boundary coordinates (100, 1120, 850, 1440) to obtain 4 functional area sub-blocks, which correspond to the construction area, construction content, shooting time, and labor information, respectively.

[0048] S22: OCR Recognition. Using the same PaddleOCR model parameters as in Example 1, image sub-blocks are recognized. For example, the recognition result of a certain photo is: construction area "K0+200-K0+300", construction content "roadbed backfilling", shooting time "2026-02-21 14:10", and labor information "12 general workers and 2 road roller drivers".

[0049] S23-S25: ​​Text Processing and Structured Data Construction. Filtering null values ​​and cleaning illegal characters, mapping the recognition results to corresponding fields, and constructing a structured data object: {"Construction Area":"K0+200-K0+300 section","Construction Content":"Roadbed Backfilling","Shooting Time":"2026-02-21 14:10","Labor Information":"12 General Workers, 2 Road Roller Drivers"}.

[0050] S3: Automatic photo renaming, archiving, and ledger generation S31-S32: Automatic renaming. Generate filenames according to the rule of "Construction Area-Construction Content-Shooting Date" such as "K0+200-K0+300 Section-Roadbed Backfill-20260221", and add incremental suffixes when there are duplicate names.

[0051] S33-S34: Hierarchical archiving. Create folders such as "K0+200-K0+300 / 2026 / 02" according to the "Construction Area / Year / Month" structure, and archive the photos to the corresponding directories.

[0052] S35: Ledger Generation. Generate a ledger data table containing the construction area, construction content, shooting time, and number of workers (14 people), and update it synchronously to the project management system.

[0053] S4: Automatic generation of construction log PDF S41: Log text generation. Generate text according to the mapping rules: "Today (2026-02-21 14:10), roadbed backfilling construction work was carried out in the K0+200-K0+300 section. A total of 14 people participated in the construction. The roadbed backfilling thickness meets the design requirements, and the compaction degree needs to be tested."

[0054] S42-S44: PDF Generation and Storage. Fill the text into the preset template, call the LaTeX compilation engine to generate a PDF file, name it "2026-02-21-K0+200-K0+300 segment-subgrade backfill", and store it in the corresponding construction log folder.

[0055] In this embodiment, the processing time for 150 photos per day is reduced from 3.5 hours to 25 minutes by manual management, the OCR recognition accuracy reaches 97.8% (effectively overcoming the effects of dust interference and insufficient lighting), the photo archiving standardization rate is 100%, and the construction log is automatically generated without manual entry, which greatly improves the information management efficiency of municipal road construction sites, realizes the traceability and queryability of the construction process, and reduces the workload of management personnel.

[0056] The two embodiments described above cover two typical construction scenarios: residential buildings and municipal roads, respectively, verifying the versatility and practicality of the method of the present invention. Image preprocessing solves the problem of low recognition accuracy caused by complex shooting environments and varying resolutions at construction sites. OCR recognition and structured data construction enable automatic extraction of watermark information. Automatic renaming, archiving, and construction log generation achieve automation and standardization of construction information management, effectively solving the problems of low efficiency, error-proneness, and disorganized archiving associated with traditional manual management. This method is suitable for the information management needs of various construction sites.

[0057] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A project management method based on automatic parsing of watermarked image information at construction sites, characterized in that, Includes the following steps: S1. Import the watermarked photo file of the construction site to be processed, and perform image preprocessing on the watermarked photo file to obtain the boundary coordinates of the standardized image and watermark text area. S2. Based on the boundary coordinates of the watermarked text region, the image is cropped, and the cropped image sub-blocks are subjected to OCR recognition to generate corresponding structured data. S3. Automatically rename and hierarchically archive the watermarked photos based on the structured data, and generate construction information ledger data; S4. Automatically generate a construction log PDF file by calling a preset document template based on the structured data.

2. The engineering management method based on automatic parsing of watermark image information at the construction site according to claim 1, characterized in that: Step S1 includes: S11. Standardize the resolution of watermarked photos at the construction site to unify the images to the preset standard resolution. S12. Perform image enhancement processing on the resolution-normalized image to improve the contrast and clarity of the text area; S13. Locate the watermark text region based on manually marked regional coordinate parameters and output the boundary coordinates of the watermark text region.

3. The engineering management method based on automatic parsing of watermark image information at the construction site according to claim 2, characterized in that: Step S11 includes: S111, Obtain the width of the original image. ,high And pixel density information, set the preset standard resolution pixel size as ; S112. Calculate the scaling ratio and width scaling ratio based on the preset standard resolution. Height scaling ratio ; S113. The image is scaled using a bicubic interpolation algorithm. The weight function of the bicubic interpolation algorithm is: ; The formula for two-dimensional interpolation merging is: ， in The pixel coordinates to be interpolated are... For decimal offset, The original pixel values ​​are the 4×4 neighborhood around the point to be interpolated. The algorithm calculation steps are as follows: First, select 4×4 grid points around the point to be interpolated, substitute them into the weight function to calculate the interpolation weights in the x and y directions respectively, perform point-by-point interpolation in the first direction, and then perform column direction interpolation to obtain the scaled target pixel value. S114. Sharpen the scaled image using a sharpening filter to reduce the loss of sharpness caused by the resolution change.

4. The engineering management method based on automatic parsing of watermark image information at the construction site according to claim 2, characterized in that: Step S12 includes: S121. Convert the image to a grayscale image. The grayscale conversion formula is: ,in These are the red, green, and blue channel values ​​of the original image pixels, respectively; The pixel values ​​of the target grayscale image; S122. Binarization is performed using an adaptive threshold segmentation method. The local threshold calculation formula for this algorithm is as follows: ,in In pixels Centered The average gray level within the neighborhood; The threshold adjustment coefficient is used; the binarization determination calculation formula is as follows: ; in The calculation steps are as follows: traverse all pixels of the grayscale image, calculate the grayscale mean value in the neighborhood of its preset size for each pixel, substitute it into the local threshold formula to obtain the adaptive threshold of the pixel, and then generate the binary image pixel value point by point according to the binarization determination rules. S123. Perform contrast enhancement and noise suppression processing on the binarized image to improve the accuracy of subsequent text recognition.

5. The engineering management method based on automatic parsing of watermark image information at the construction site according to claim 2, characterized in that: Step S13 includes: S131. During the system initialization phase, sample watermark photos are marked with regions through manual interaction, and the boundary coordinates of the watermark text region are recorded. ; S132. Save the boundary coordinates as clipping region parameters in the configuration file; S133. During the batch processing stage, the cropping region parameters are read, and the region coordinates are proportionally adapted according to the current image size. The adaptation calculation formula is as follows: ,in , which is the scaling ratio of the width and height of the current image compared to the sample image; To adapt to the target boundary coordinates; S134. Based on the adapted region coordinates, crop and locate the image, and output the boundary coordinates of the watermark text region.

6. The engineering management method based on automatic parsing of watermark image information at the construction site according to claim 1, characterized in that: Step S2 includes: S21. Based on the boundary coordinates of the watermark text region output by S1, crop the image to obtain image sub-blocks with multiple fixed functional regions. S22. The image sub-blocks are input into an OCR text recognition model for recognition. The OCR text recognition model is composed of the DBNet text detection model and the CRNN text recognition model in the PaddleOCR framework; wherein the differentiable binarization calculation formula of DBNet is: , This is the magnification factor; These are the pixel values ​​of the probability map. The threshold image pixel values; the CTC loss calculation formula for CRNN is: ; in To map to the target sequence The set of all paths The path probability is calculated using the PaddleOCR model algorithm. The DBNet text detection network is used to detect text regions in image sub-blocks. After feature fusion and bi-branch prediction, a text region probability map and a threshold map are obtained. Substituting these into the differentiable binarization formula generates an approximate binary map. Then, contour extraction and NMS non-maximum suppression are used to obtain accurate text detection boxes. The text image within the detection box is then input into a CRNN. ​​After CNN convolutional feature extraction, Bi-LSTM bidirectional recurrent sequence modeling, and Softmax classification, the sequence alignment and character recognition are completed by substituting into the CTC loss formula, and the text character sequence is output. S23. Extract text and filter null values ​​from the OCR recognition results, and concatenate multiple lines of text within the same area. S24. According to the preset area function mapping rules, the recognition results are mapped to the construction area field, construction content field, shooting time field and labor information field respectively. S25. Clean up and standardize the text of each field to create a structured data object.

7. The engineering management method based on automatic parsing of watermark image information at the construction site according to claim 1, characterized in that: Step S3 includes: S31. Based on the construction area, construction content, and shooting time fields in the structured data, automatically rename the watermarked photos according to the naming rule of "construction area-construction content-shooting date". S32. When a newly generated filename contains duplicates, an incrementing suffix is ​​added to avoid duplicate names; the formula for duplicate name correction is as follows: Where N is the original renaming result, n is a positive integer incrementing, n=1,2,3..., and N′ is the final unique filename. S33. Based on the construction area and shooting time fields, automatically create multi-level folders according to the hierarchical directory structure of "construction area / year / month"; S34. Archive the renamed watermarked photos to the corresponding folder; S35. Generate a construction ledger data table containing fields for construction area, construction content, shooting time, and number of laborers.

8. The engineering management method based on automatic parsing of watermark image information at the construction site according to claim 1, characterized in that: Step S4 includes: S41. Generate construction log text data based on structured data; convert the structured data into log text content according to preset field mapping rules, with the mapping operation relationship as follows: ,in These fields are: construction area, construction content, shooting time, and labor information. This is a predefined mapping function from fields to log text. S42. Fill the construction log text data into the preset document template variable field; S43. Use the LaTeX compilation engine to generate PDF format construction log files; S44. Name the generated construction log PDF files according to the construction time and store them in the corresponding construction log folder.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the engineering management method based on the automatic parsing of watermark image information at the construction site as described in any one of claims 1 to 8.

Images