Project tracking management method based on data analysis

Abstract

The invention relates to the technical field of project management, and particularly discloses a project tracking management method based on data analysis, which comprises the following steps of: establishing a first character recognition model and a second character recognition model, and deploying the first character recognition model and the second character recognition model at a cloud end and a local end respectively; obtaining a minimum enclosing rectangle of a target character in the project report, obtaining a sub-part based on the minimum enclosing rectangle, and determining the grid property of the grid based on the sub-part; abnormal characters and normal characters are determined according to the grid properties, and the abnormal characters and the normal characters are recognized through a first character recognition model and a second character recognition model respectively; and combining the recognition result of the first character recognition model and the recognition result of the second character recognition model into a text and then displaying the text. According to the invention, the accuracy of project tracking management is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of project management, and in particular to a project tracking management method based on data analysis. Background Art

[0002] Project tracking is a crucial aspect of project management, ensuring real-time visibility into project execution. By establishing milestones and critical paths, we continuously collect progress data, compare plans with actual results, and promptly identify and implement corrective measures. Furthermore, through visualization tools and reports, we ensure transparent communication, ensuring team members have a clear understanding of task completion status and next steps.

[0003] In current construction management practices, frontline workers are required to handwrite daily information such as completion volume, quality inspections, and resource usage in logs or reports. These records are then photographed or scanned to create electronic files and uploaded to backend systems for review by management. However, varying handwriting habits and site conditions often result in illegible and sloppy handwriting, making it difficult for managers to accurately and timely assess construction progress and quality through electronic files, thus hindering efficient management decision-making. Summary of the Invention

[0004] The purpose of this invention is to provide a project tracking management method based on data analysis to solve the following technical problems:

[0005] In current construction management practices, frontline workers are required to handwrite daily information such as completion volume, quality inspections, and resource usage in logs or reports. These records are then photographed or scanned to create electronic files and uploaded to backend systems for review by management. However, varying handwriting habits and site conditions often result in illegible and sloppy handwriting, making it difficult for managers to accurately and timely assess construction progress and quality through electronic files, thus hindering efficient management decision-making.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A project tracking management method based on data analysis includes the following steps:

[0008] Establishing a first character recognition model and a second character recognition model, wherein the number of layers of the first character recognition model is greater than the number of layers of the second character recognition model, and the first character recognition model and the second character recognition model are deployed in the cloud and locally, respectively;

[0009] Obtaining a project report uploaded by a construction worker, obtaining a target character based on the project report, obtaining a minimum bounding rectangle of the target character, dividing the minimum bounding rectangle into a grid to obtain a plurality of grids, recording the target character in a single grid portion as a sub-portion, and determining a grid property of the grid based on the sub-portion, wherein the grid property includes normal and abnormal.

[0010] The target character corresponding to the grid with abnormal grid properties is regarded as an abnormal character, and the abnormal character is sent to the first character recognition model for recognition; the target character corresponding to the grid with normal grid properties is regarded as a normal character, and the normal character is sent to the second character recognition model for recognition;

[0011] The recognition results of the first character recognition model and the recognition results of the second character recognition model are combined into text and then displayed.

[0012] As a further solution of the present invention: the process of obtaining the target character includes:

[0013] Based on contour detection, characters in the project report are segmented and recorded as initial characters. The initial characters are enlarged to obtain target characters. The enlargement ratio is a preset value and is greater than 2.

[0014] As a further aspect of the present invention, the process of determining the grid properties of the grid includes:

[0015] Obtaining similarities between subpart i and radicals in a preset radical list, generating a similarity set sum Xijh={Xi1, Xi2, ..., Xim}, where Xim represents the similarity between the subpart i and the mth radical in the radical list;

[0016] Get the maximum similarity Xi'=max(Xijh) and calculate the total similarity n represents the total number of the grids;

[0017] When the total similarity is less than a preset total similarity threshold, the grid property of the corresponding grid is recorded as abnormal; when the total similarity is greater than or equal to the preset total similarity threshold, the grid property of the corresponding grid is recorded as normal.

[0018] As a further aspect of the present invention, the process of determining the grid properties of the grid further includes:

[0019] Recording the radical corresponding to the maximum similarity Xi' as the target part of the subpart i, obtaining the target part corresponding to the single target character, and judging whether the constraint is satisfied. If not, recording the grid property of the corresponding grid as abnormal; if so, recording the grid property of the corresponding grid as normal;

[0020] The constraint is that a character obtained by combining the target parts corresponding to a single target character according to the positions of the corresponding sub-parts in the minimum circumscribed rectangle belongs to a preset character list.

[0021] As a further solution of the present invention: when the maximum similarity Xi' is less than or equal to a preset maximum similarity threshold, the grid property of the corresponding grid is recorded as abnormal.

[0022] As a further solution of the present invention, the process of combining the recognition results of the first character recognition model and the recognition results of the second character recognition model into text and then displaying the text includes:

[0023] The recognition results of the first character recognition model and the recognition results of the second character recognition model are sorted according to the order in which the corresponding initial characters appear in the project report to obtain a combined text, and the text is visually displayed.

[0024] As a further solution of the present invention, before visually displaying the text, the following steps are further included:

[0025] Obtain the perplexity of a single sentence in the text, count the number of sentences whose perplexity is greater than a preset threshold, and send a prompt message to the construction worker if the number exceeds the preset threshold.

[0026] As a further solution of the present invention: the perplexity is obtained based on a pre-trained language model.

[0027] The beneficial effects of the present invention are as follows:

[0028] 1) This invention uses a hierarchical character recognition architecture of "local lightweight - cloud deep" to retain characters that are normal after being determined by the grid for immediate processing at the terminal, while uploading suspected abnormal characters to the cloud for detailed recognition. This not only maintains overall recognition accuracy, but also avoids the communication and computing congestion caused by remote reasoning of all images, ensuring smooth data entry and immediate feedback at the construction site. Structured data can be continuously transmitted back to the backend, allowing managers to grasp progress and quality in real time, eliminating the time lag and subjective errors caused by manual data entry. At the same time, small-scale updates of the local model can be completed without downtime, reducing equipment maintenance and network dependence.

[0029] 2) Through the calculation of radical similarity and the grid-level anomaly determination mechanism, this solution can accurately separate and locate abnormal stroke blocks in the case of connected, blurred or partially stained handwriting, and then divert them to high-precision model processing before formal recognition. This fine-grained filtering avoids the misjudgment of the entire image being rejected as a whole due to local shadows or creases, significantly enhancing the system's robustness to noise and missing strokes. Spatially constrained reorganization of sub-parts can automatically remove pseudo-characters caused by light spots and wrinkles, ensuring that the exported text maintains high consistency with the original writing semantics, providing a reliable data foundation for progress statistics and quality warnings, and further supporting the precise scheduling of multi-process collaborative management;

[0030] 3) After character recognition and sorting are complete, the system uses a pre-trained language model to calculate sentence perplexity and monitor the readability and logical integrity of the report text in real time. When the number of sentences with abnormal perplexity exceeds the threshold, the platform immediately sends a prompt to the construction worker, guiding them to revise problems such as irregular handwriting and missing content before taking photos on-site or uploading them, thereby preventing erroneous data from entering the back-end process. The closed-loop feedback loop formed in this way provides management with clearer and more consistent information, reducing the time spent on repeated communication and confirmation. At the same time, automated quality gating reduces the pressure on manual back-end review, allowing resources to be focused on abnormal work point analysis and decision support, thereby improving the overall level of refinement and intelligence of project tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention will be further described below with reference to the accompanying drawings.

[0032] Figure 1 It is a flow chart of a project tracking management method based on data analysis of the present invention. DETAILED DESCRIPTION

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0034] See also Figure 1 As shown, the present invention is a project tracking management method based on data analysis, comprising the following steps:

[0035] Establishing a first character recognition model and a second character recognition model, wherein the number of layers of the first character recognition model is greater than the number of layers of the second character recognition model, and the first character recognition model and the second character recognition model are deployed in the cloud and locally, respectively;

[0036] It should be noted that when building the character recognition model required for project tracking and management, we first collect the handwritten daily reports and quality record photos uploaded from the construction site, perform careful cropping and annotation operations on these photos, separate each Chinese character, and assign it an accurate label. In order to enhance the adaptability of the model to complex actual scenes, a series of data expansion methods are used, such as random blurring, brightness drift, stain occlusion, perspective distortion and other processing methods, to simulate the common light spots, dust, creases and other noise conditions on the construction site, so that the model can be fully exposed to similar complex scenes during the training stage; after that, the training of the first character recognition model is carried out in the cloud. This model adopts a deep convolution-attention hybrid network architecture. Its network depth and parameter quantity are significantly improved compared to the local model to be built later. By leveraging the powerful computing power and ample memory of the cloud server, this model can extract characters on a single character image. After training, this highly accurate first character recognition model, which uses richer stroke, structural, and contextual features, is deployed to a cloud inference cluster using containerization technology to provide accurate services for subsequent character recognition tasks. Simultaneously, based on the same previously labeled data, a second character recognition model is constructed using pruning and distillation technology. This local model is lightweight and can run in real time on the CPU of an ordinary tablet or mobile phone by compressing channels and reducing the number of layers, as well as applying optimization methods such as tensor quantization and operator fusion. Ultimately, this lightweight second character recognition model is packaged into the on-site inspection app.

[0037] Obtaining a project report uploaded by a construction worker, obtaining a target character based on the project report, obtaining a minimum bounding rectangle of the target character, dividing the minimum bounding rectangle into a grid to obtain a plurality of grids, recording the target character in a single grid portion as a sub-portion, and determining a grid property of the grid based on the sub-portion, wherein the grid property includes normal and abnormal.

[0038] In a preferred embodiment of the present invention, the process of obtaining the target character includes:

[0039] Segmenting characters in the project report based on contour detection, recording them as initial characters, and amplifying the initial characters to obtain target characters, wherein the amplification ratio is a preset value and is greater than 2;

[0040] In another preferred embodiment of the present invention, the process of determining the grid properties of the grid includes:

[0041] Obtaining similarities between subpart i and radicals in a preset radical list, generating a similarity set sum Xijh={Xi1, Xi2, ..., Xim}, where Xim represents the similarity between the subpart i and the mth radical in the radical list;

[0042] Obtain the maximum similarity Xi' = max(Xijh) and calculate the total similarity n represents the total number of the grids;

[0043] When the total similarity is less than the preset total similarity threshold, mark the grid property of the corresponding grid as abnormal; when the total similarity is greater than or equal to the preset total similarity threshold, mark the grid property of the corresponding grid as normal;

[0044] It should be noted that after the construction worker uploads the project report image through the mobile terminal, the system performs preprocessing operations on the image, including denoising, binarization, and connected component analysis, to highlight the text content. Subsequently, with the help of the contour detection operator, the system precisely outlines each ink blob, and determines the closed contours that can be clearly separated from the background as individual characters and temporarily stores them as initial characters. To improve the character resolution for more detailed analysis of the strokes later, the system uses an equidistant scaling function to magnify each initial character by a certain multiple while always maintaining the original aspect ratio to avoid character deformation; after obtaining the magnified character, the system calculates its minimum bounding rectangle, just like customizing a bounding box for the character. Then, this rectangle is evenly divided into multiple grids, just like covering the character with a transparent chessboard. The stroke pixels intersecting with the character in each grid are regarded as a sub - part; for each sub - part, the system extracts its histogram of oriented gradients and compares it one by one with the preset radical template library to calculate the similarity vector. For example, if the shape of a certain sub - part highly fits the local contour of a radical such as "氵" or "口", then the similarity value of this sub - part with the corresponding template will be at a relatively high level in the vector. The radical corresponding to the maximum value in the vector is the closest shape of this sub - part. Add up the maximum similarities of all sub - parts, and the sum obtained is the total similarity of this grid. If this total similarity is lower than the preset empirical threshold, or there are obvious stains resulting in the sub - part being unable to form a reliable match with any radical, the system will mark this grid as abnormal; otherwise, it is determined to be normal. In this way, before the characters enter the formal recognition stage, the system can accurately screen out the areas that may affect recognition due to dirt, warping, or scribbled writing, and hand them over to the high - precision model for processing, while the clear and complete areas can be directly recognized by the lightweight model quickly, accelerating the overall processing flow while ensuring the reliability of the final recognition result;

[0045] In a preferred case of this embodiment, the process of determining the grid property of the grid further includes:

[0046] Denote the radical corresponding to the maximum similarity Xi' as the target part of the sub - part i, obtain the target part corresponding to a single target character, and determine whether the constraint is satisfied. If not, record the grid property of the corresponding grid as abnormal; if so, record the grid property of the corresponding grid as normal.

[0047] The constraint is that the character formed by combining the target parts corresponding to a single target character according to the positions of the corresponding sub - parts in the minimum bounding rectangle belongs to a preset character list.

[0048] It should be noted that when the maximum similarity Xi' is less than or equal to the preset maximum similarity threshold, record the grid property of the corresponding grid as abnormal.

[0049] It can be understood that after comparing the similarity between each sub - part and the radical template, the system will record the radical with the highest matching score as the target part of the sub - part. Then, collect the target parts of all sub - parts within the same target character and recombine them in the order from left to right and from top to bottom according to their positions in the minimum bounding rectangle to form a "candidate character". For example, if the target part corresponding to the left - hand grid is determined to be "氵" and the target part corresponding to the right - hand grid is "青", and their positional relationship conforms to the normal Chinese character writing specification, and the combined character "清" happens to exist in the pre - set list of common Chinese characters, this indicates that the spatial relationship of each stroke of the target character is reasonable and the structure is complete, and the system will mark the relevant grid as normal. On the contrary, if the spatial distribution of the sub - parts does not conform to the standard glyph, such as "氵" appears on the right, or the parts overlap vertically or horizontally, resulting in the combined character not corresponding to any character in the pre - set Chinese character list, the system will determine that the grid is abnormal and transfer it to the subsequent high - precision recognition process for processing. In addition, if the maximum similarity of a certain sub - part is lower than the preset reliability standard, even if the character formed by combining the remaining sub - parts meets the requirements, the system will directly determine that the grid is abnormal to avoid the influence of invalid characters caused by stains, creases or overly scribbled writing on the final recognition result.

[0050] Regard the target character corresponding to the grid with abnormal grid property as an abnormal character, and send the abnormal character to the first character recognition model for recognition; regard the target character corresponding to the grid with normal grid property as a normal character, and send the normal character to the second character recognition model for recognition.

[0051] Combine the recognition results of the first character recognition model and the second character recognition model into text and then display it.

[0052] The process of combining the recognition result of the first character recognition model and the recognition result of the second character recognition model into text and then displaying the text includes:

[0053] Sorting the recognition results of the first character recognition model and the recognition results of the second character recognition model according to the order in which the corresponding initial characters appear in the project report to obtain a combined text, and visually displaying the text;

[0054] It is worth noting that the process of combining the recognition result of the first character recognition model and the recognition result of the second character recognition model into text and then displaying the text includes:

[0055] Sorting the recognition results of the first character recognition model and the recognition results of the second character recognition model according to the order in which the corresponding initial characters appear in the project report to obtain a combined text, and visually displaying the text;

[0056] It should be noted that in the process of project tracking and management, the construction site environment is complex, and character images are easily affected by factors such as contamination and illegible writing. The present invention adopts a character diversion recognition strategy to send the characters judged as abnormal by the grid to the high-precision first character recognition model deployed in the cloud. Because of its powerful feature extraction and analysis capabilities, it can accurately process these complex characters; and the normal characters are handed over to the local lightweight second character recognition model for processing, which can achieve rapid on-site recognition. This diversion mechanism avoids the network congestion and processing delays caused by uploading all images to the cloud. Construction workers can quickly complete most character entry on site, and the operation is smooth and unstuck. At the same time, the high-precision model in the cloud only needs to focus on processing the real difficult characters, so that computing resources are used efficiently, which greatly improves the recognition accuracy in complex scenarios. Ultimately, the recognition results of the two models are integrated into a complete text in the order of the original report. The management backend can quickly obtain accurate structured data. Whether it is project progress statistics, quality traceability, or decision analysis, it has solid data support. This effectively avoids the inefficiency and misjudgment problems that may be caused by a single recognition path, comprehensively enhances the real-time nature of project tracking management and the credibility of data, and provides a strong guarantee for the smooth progress of the project.

[0057] It should be noted that before visualizing the text, the following steps are also included:

[0058] Obtain the perplexity of a single sentence in the text, count the number of sentences whose perplexity is greater than a preset threshold, and send a prompt message to the construction worker if the number exceeds the preset threshold;

[0059] It should be noted that the perplexity is obtained based on a pre-trained language model;

[0060] It is worth noting that after the recognition results are integrated into complete text, the sentence perplexity detection mechanism is introduced. This is equivalent to adding an intelligent quality checkpoint before project information enters the management end. With the help of pre-trained language models such as "Chinese-GPT" that have been fully learned and trained on massive text corpora, each sentence in the text is deeply semantically analyzed, the corresponding perplexity value is calculated, and the number of sentences with perplexity exceeding the preset standard is counted. Once the number of abnormal sentences exceeds the set threshold, the system will immediately send a prompt message to the construction worker, guiding him to promptly check, supplement or correct the ambiguous or missing content on site. First, it can pre-identify and filter text readability defects caused by typos, incoherent sentences, incomplete information, etc., effectively preventing erroneous or ambiguous data from entering the visualization report, and fundamentally preventing decision-making bias caused by data quality issues. Second, by realizing the immediate discovery and correction of problems on the client side, a closed loop of data quality control is formed, which greatly reduces the workload of manual review in the backend and avoids tedious repeated communication, ensuring that the text content presented to management is both accurate and clear while maintaining a high degree of timeliness. Ultimately, it provides reliable data support for key management links such as project progress assessment and quality warning, and effectively promotes the realization of project tracking management goals.

[0061] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A project tracking management method based on data analysis, characterized in that: The following steps are involved: Establishing a first character recognition model and a second character recognition model, wherein the number of layers of the first character recognition model is greater than the number of layers of the second character recognition model, and the first character recognition model and the second character recognition model are deployed in the cloud and locally, respectively; Obtaining a project report uploaded by a construction worker, obtaining a target character based on the project report, obtaining a minimum bounding rectangle of the target character, dividing the minimum bounding rectangle into a grid to obtain a plurality of grids, recording the target character in a single grid portion as a sub-portion, and determining a grid property of the grid based on the sub-portion, wherein the grid property includes normal and abnormal. The target character corresponding to the grid with abnormal grid properties is regarded as an abnormal character, and the abnormal character is sent to the first character recognition model for recognition; the target character corresponding to the grid with normal grid properties is regarded as a normal character, and the normal character is sent to the second character recognition model for recognition; The recognition results of the first character recognition model and the recognition results of the second character recognition model are combined into text and then displayed.

2. A project tracking management method based on data analysis according to claim 1, characterized in that: The process of obtaining the target character includes: Based on contour detection, characters in the project report are segmented and recorded as initial characters. The initial characters are enlarged to obtain target characters. The enlargement ratio is a preset value and is greater than 2.

3. The project tracking management method based on data analysis according to claim 1, characterized in that: The process of determining the grid properties of the grid includes: Obtaining similarities between subpart i and radicals in a preset radical list, generating a similarity set sum Xijh={Xi1, Xi2, ..., Xim}, where Xim represents the similarity between the subpart i and the mth radical in the radical list; Get the maximum similarity Xi'=max(Xijh) and calculate the total similarity n represents the total number of the grids; When the total similarity is less than a preset total similarity threshold, the grid property of the corresponding grid is recorded as abnormal; when the total similarity is greater than or equal to the preset total similarity threshold, the grid property of the corresponding grid is recorded as normal.

4. A project tracking management method based on data analysis according to claim 3, characterized in that: The process of determining the grid properties of the grid further includes: Recording the radical corresponding to the maximum similarity Xi' as the target part of the subpart i, obtaining the target part corresponding to the single target character, and judging whether the constraint is satisfied. If not, recording the grid property of the corresponding grid as abnormal; if so, recording the grid property of the corresponding grid as normal; The constraint is that a character obtained by combining the target parts corresponding to a single target character according to the positions of the corresponding sub-parts in the minimum circumscribed rectangle belongs to a preset character list.

5. The project tracking management method based on data analysis according to claim 3 is characterized in that: When the maximum similarity Xi' is less than or equal to a preset maximum similarity threshold, the grid property of the corresponding grid is recorded as abnormal.

6. A project tracking management method based on data analysis according to claim 2, characterized in that: The process of combining the recognition result of the first character recognition model and the recognition result of the second character recognition model into text and then displaying the text includes: The recognition results of the first character recognition model and the recognition results of the second character recognition model are sorted according to the order in which the corresponding initial characters appear in the project report to obtain a combined text, and the text is visually displayed.

7. A project tracking management method based on data analysis according to claim 6, characterized in that: Before visualizing the text, the following steps are also included: Obtain the perplexity of a single sentence in the text, count the number of sentences whose perplexity is greater than a preset threshold, and send a prompt message to the construction worker if the number exceeds the preset threshold.

8. A project tracking management method based on data analysis according to claim 7, characterized in that: The perplexity is obtained based on a pre-trained language model.

Images