Defect annotation-based decoration engineering specification clause recommendation method and system

Abstract

The present application relates to the technical field of building decoration engineering quality management, and discloses a decoration engineering specification recommendation method and system based on defect labeling. The present application realizes the structured association of the inspection pictures, the inspection context information and the defect record object containing the labeling coordinates, the defect label and the severity level, converts the defect data from the scattered pictures and the text description into the standardized data that can be efficiently stored, retrieved and processed by the computer system, solves the problem that the defect record is difficult to deposit and reuse in the existing inspection, realizes the automatic recommendation of the specification through the double-path recall mechanism combining the defect label mapping recall with the extended retrieval based on the field description and the inspection item content, and the multi-factor weighted scoring and sorting method, reduces the workload of the manual review of the standard files by the inspection personnel, and improves the specification query efficiency.

Description

Technical Field

[0001] This invention relates to the field of quality management technology for building decoration and renovation projects, and in particular to a method and system for recommending standard clauses for decoration and renovation projects based on defect labeling. Background Technology

[0002] Decoration and renovation construction sites are characterized by numerous work processes, detailed defect types, scattered quality issues, and complex standards and regulations. Current on-site quality inspections typically rely on inspectors taking photos, manually describing problems, and manually consulting relevant regulations, which has the following shortcomings:

[0003] 1. There is usually a lack of structured annotations between on-site photos and defect areas, making it difficult to subsequently summarize them into a unified defect sample.

[0004] 2. Defect types, severity, location, and on-site descriptions are often recorded in a scattered manner, making it difficult to form a defect record object that can be directly reused.

[0005] 3. After inspectors discover defects, they still need to manually check national standards, industry standards, or enterprise standards, which is inefficient and the cited clauses for the same type of defect are inconsistent among different personnel.

[0006] 4. Even if some standard clauses are pre-set in the system, it can usually only perform simple keyword searches or fixed mappings, and cannot combine defect labels, on-site instructions, inspection context and standard validity to make more accurate clause recommendations.

[0007] 5. There is usually a lack of a unified data organization method between the defect annotation results and the recommended results of the standard provisions, making it difficult to form a stable and interpretable basis for the recommendation of provisions.

[0008] In existing technologies, one type of solution focuses on pure image defect detection, emphasizing the identification of defect location or defect category; the other type focuses on text retrieval or text recommendation, emphasizing the matching of legal clauses from text. For the inspection scenario of decoration and renovation projects, there is still a lack of a technical solution that can take defect annotation results as the entry point and unify and link defect area, defect label, defect severity, on-site description, and recommended standard clauses.

[0009] Therefore, a method and system for recommending standard provisions for decoration and renovation engineering inspection scenarios is needed to achieve structured defect labeling, consistent standard provision recommendations, and interpretable recommendation basis. Summary of the Invention

[0010] To overcome the shortcomings of existing recommendations for standard provisions in decoration and renovation engineering inspection scenarios, such as the lack of a method that uses defect annotation results as an entry point and integrates defect areas, defect labels, defect severity, on-site descriptions, and standard provision recommendations, this invention provides a method and system for recommending standard provisions for decoration and renovation engineering based on defect annotation.

[0011] In a first aspect, the present invention provides a method for recommending clauses in decoration and renovation engineering specifications based on defect annotation, including: Receive inspection images sent by the inspection terminal, as well as the inspection item identifiers and construction location information associated with the inspection images; The inspection terminal receives a defect record object sent by the inspection terminal. The defect record object is generated by the inspection terminal based on the defect area selected and marked by the user on the inspection image. The defect record object includes at least the defect image path, the coordinates of the marked area, the defect label, the defect severity level, and the on-site description. Based on the defect label identifier, a set of candidate standard clauses is recalled from the preset mapping relationship between defect labels and standard clauses, and the set of candidate standard clauses is expanded or filtered based on at least one of the following: on-site instructions, content of inspection items, construction location information, and process information. A recommendation score is calculated for each candidate specification clause in the candidate specification clause set. The recommendation score is determined based on at least two of the following factors: the basic mapping weight between the defect label and the candidate specification clause; the text matching result between the on-site description in the defect record object and the content of the candidate specification clause; the consistency result between the content of the inspection item and the scope of application of the candidate specification clause; the validity status or version information of the candidate specification clause; and the priority correction value corresponding to the defect severity level. Based on the recommended scores, the candidate standard provisions in the candidate standard provision set are sorted, and the recommended standard provisions are output to the inspection terminal. The standard number, standard name, provision content, and recommendation basis of the recommended standard provisions are also output for the inspection terminal to display.

[0012] According to one specific implementation, in the above recommended method, the defect record object further includes a note and a generation time.

[0013] According to a specific implementation, in the above-mentioned recommended method, the defect label identifier corresponds to the leaf label node in the tree-shaped defect label system, and the parent node in the tree-shaped defect label system is used for coarse-grained recall of candidate specification clauses; the defect label is pre-associated with the construction location and construction procedure.

[0014] According to a specific implementation, in the above-mentioned recommended method, the step of recalling a set of candidate specification clauses from a preset mapping relationship between defect tags and specification clauses based on the defect tag identifier includes: Based on the defect label identifier, obtain the first set of candidate clauses from the defect label and specification clause mapping table; Based on at least one of the keywords in the site description, the contents of the inspection items, and the construction location information, a second set of candidate clauses is retrieved from the standard library. The first candidate clause set and the second candidate clause set are merged and deduplicated. When the first candidate text set is empty or the number is less than a preset threshold, an extended recall based on keyword retrieval, semantic retrieval, or a combination of both is performed.

[0015] According to a specific implementation, in the above recommendation method, the text matching result includes keyword matching degree and semantic similarity, and the semantic similarity is calculated based on the similarity between the text vector of the defect record object and the text vector of the candidate specification clause.

[0016] According to one specific implementation, in the above recommendation method, the recommendation score is calculated according to the following formula:

[0017] in, This represents the recommendation score for the i-th candidate specification clause. This represents the basic mapping weight between the defect label and the i-th candidate specification clause. This represents the text matching score between the stated on-site description and the content of the i-th candidate specification clause. The score represents the contextual consistency between the content, construction location information, or process information of the inspection item and the scope of application of the i-th candidate specification clause. This represents the correction value corresponding to the validity status or version information of the i-th candidate specification clause. This indicates the priority adjustment value corresponding to the severity level of the defect. , , , , These are the weighting coefficients, and the sum of the five factors equals 1.

[0018] According to a specific implementation method, in the above-mentioned recommended method, the context consistency score is calculated by weighting the content matching results of the inspection items, the construction parts or procedures matching results, and the construction location matching results; the priority correction value corresponding to the severity level of the defect is mapped to the corresponding value in ascending order of severity; the correction value corresponding to the validity status or version information of the candidate specification clause is the first value when the clause is valid, the second value when the clause is valid but not a priority version, and the third value or the clause is removed from the candidate specification clause set when the clause is invalid.

[0019] According to a specific implementation method, in the above recommendation method, when the candidate specification clause set is empty, the text retrieval based on the content of the on-site instructions and the inspection items is downgraded; when the recommendation scores of all candidate specification clauses are lower than a preset threshold, a manual review prompt is output and at least one candidate specification clause with the highest score is displayed; when there are multiple versions of the same standard number, the candidate specification clause with the most valid status and the latest version is retained.

[0020] Secondly, the present invention provides a recommendation system for decoration and renovation engineering specification clauses based on defect annotation, including an inspection terminal and a server; The inspection terminal includes: The image acquisition module is used to acquire inspection images, associate the inspection images with inspection item identifiers and construction location information, and send the inspection images to the server. The defect annotation module is used to receive the user's box selection annotation operation on the inspection image, generate defect annotation area data, and generate a defect record object by combining the defect label, defect severity level and on-site description, and send the defect record object to the server. The results display module is used to receive and display the recommended specification clauses returned by the server, along with their corresponding standard numbers, standard names, clause content, and recommendation basis. The server includes: The candidate clause recall module is used to receive the inspection images and the defect record objects, recall a set of candidate standard clauses from the mapping relationship between the preset defect tags and standard clauses according to the defect tag identifier, and expand or filter the set of candidate standard clauses according to at least one of the following: on-site description, content of inspection items, construction location information and process information. The scoring and ranking module is used to calculate a recommendation score for each candidate specification clause in the candidate specification clause set. The recommendation score is determined based on at least two of the following: the basic mapping weight between the defect tag and the candidate specification clause; the text matching result between the on-site description and the content of the candidate specification clause; the consistency result between the content of the inspection item and the applicable scope of the candidate specification clause; the effective status or version information of the candidate specification clause; and the priority correction value corresponding to the defect severity level. The module then ranks each candidate specification clause in the candidate specification clause set according to the recommendation score. The recommended output module is used to return the sorted recommended standard provisions and their corresponding standard numbers, standard names, provisions, and recommendation basis to the inspection terminal.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention structures and associates inspection images, inspection context information, and defect record objects containing labeled coordinates, defect tags, and severity levels. This transforms scattered image and text descriptions into standardized data that can be efficiently stored, retrieved, and processed by computer systems, solving the problem of difficulty in accumulating and reusing defect records in existing inspections. Through a dual-path recall mechanism combining defect tag mapping-based recall with extended retrieval based on on-site descriptions and inspection item content, along with a multi-factor weighted scoring and ranking method, it achieves automatic recommendation of standard clauses. This reduces the workload of inspection personnel manually searching standard documents, improves clause retrieval efficiency, and ensures consistency in the clauses cited by different personnel and in different projects for similar defects. By synchronously outputting the standard number, standard name, clause content, and recommendation basis, the recommendation results are interpretable and easy for inspection personnel to confirm on-site. Attached Figure Description

[0022] Figure 1 A flowchart illustrating a method for recommending standard clauses in decoration and renovation engineering based on defect annotation, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the candidate clause sorting process provided in an embodiment of the present invention. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to specific embodiments. However, this should not be construed as limiting the scope of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0024] Unless otherwise specified, the use of terms such as "first," "second," and "third" in the description of specific embodiments of the present invention is merely for distinguishing descriptions of identical or similar components and should not be construed as emphasizing or implying the relative importance of a particular component.

[0025] The present invention provides a method and system for recommending standard clauses for decoration and renovation engineering based on defect annotation, applicable to quality inspection scenarios at decoration and renovation construction sites. In this scenario, quality inspectors use an inspection terminal to take photos of the construction site, and mark the areas of defects found in the photos. The system automatically recommends applicable standard clauses based on the defect information. Inspectors can directly use the recommended results to explain quality problems, prepare rectification notices, or write inspection reports.

[0026] The following describes in detail, with reference to the accompanying drawings, the specific implementation methods and systems for recommending decoration and renovation engineering specification clauses based on defect annotation provided by the present invention.

[0027] Please refer to Figure 1 This illustrates a flowchart of a method for recommending standard clauses in decoration and renovation engineering based on defect annotation, provided by an embodiment of the present invention. Figure 1 As shown, the method includes the following steps.

[0028] Step 1: Receive the inspection images sent by the inspection terminal, along with the inspection item identifiers and construction location information associated with the inspection images.

[0029] For example, a communication connection is pre-established between the server and the inspection terminal before performing this step. This communication connection can be established through a wireless network (such as a 4G / 5G mobile communication network or a Wi-Fi network), and the inspection terminal and the server exchange data using HTTP or HTTPS protocols.

[0030] For example, after arriving at the quality inspection point at the decoration and renovation project site, the inspection personnel initiate the inspection task through the inspection application (APP) installed on the inspection terminal. The inspection terminal uses its built-in camera to capture inspection images of the current construction area. The inspection images can be in JPEG, PNG, or other common image formats. After the acquisition is complete, the inspection terminal displays the inspection image in the inspection application interface and prompts the inspection personnel to enter or select contextual information associated with the inspection image.

[0031] Specifically, the inspection terminal associates inspection images with the following information: inspection item identifier and construction location information. The inspection item identifier is a unique number pre-defined in the inspection task; for example, the identifier for "wall tile flatness inspection" is "INSP-001". The construction location information is a description of the specific construction location where the defect is found, such as "east wall of the living room" or "south wall of the bathroom". Optionally, the inspection terminal can further associate inspection images with the inspection task identifier, project name, collection time, and inspection personnel information. The inspection task identifier is a unique number for this inspection task, used to aggregate all inspection data under the same task on the server side. The project name is the name of the current decoration and renovation project, such as "renovation project of a residential community". The collection time is the timestamp of the inspection terminal collecting the inspection image. The inspection personnel information is the identity information of the personnel performing this inspection operation, such as the inspector's employee number or name.

[0032] After completing the above association, the inspection terminal encapsulates the inspection image, along with its associated inspection item identifiers, construction location information (and optionally, inspection task identifiers, project name, collection time, and inspection personnel information), into a data packet and sends it to the server via a wireless network. The server receives the data packet through its communication interface, parses it, and extracts the inspection image and its associated inspection item identifiers, construction location information, and other fields. The server stores the received data in its database or file system for use in subsequent steps.

[0033] By associating inspection images with structured inspection context information in the above manner, subsequent defect records and clause recommendations can be fully supported by scenario information, avoiding inspection images becoming isolated data silos and laying the foundation for the structured accumulation of defect data.

[0034] Step 2: Receive the defect record object sent by the inspection terminal.

[0035] For example, after the inspection terminal sends the inspection image to the server, the inspection terminal's inspection application displays the image on the screen. The inspection personnel then use touch controls to select and mark defect areas on the inspection image. Specifically, the inspection personnel drag their finger or stylus across the image to form a rectangle, and the area enclosed by this rectangle is the defect area. The inspection terminal generates defect marking area data based on this selection and marking operation.

[0036] After generating the defect labeling area data, the inspection terminal prompts the inspectors to select a defect label, select or enter the defect severity level, and enter a site description. Defect labels can be selected from a pre-configured list of defect labels. Defect severity levels can include, for example, four levels: "low," "medium," "high," and "critical." The site description is a supplementary written description of the defect by the inspectors, such as "Approximately 0.12 square meters of hollow wall tiles, with uneven gap width."

[0037] The inspection terminal generates a defect record object based on the defect annotation area data, defect label, defect severity level, and on-site description. This defect record object is a structured data object that includes at least the following fields: defect image path (i.e., the storage path of the inspection image on the server or inspection terminal), annotation area coordinates (i.e., the position information of the rectangle selected by the inspector on the image), defect label identifier (i.e., the unique number of the selected defect label), and defect severity level (i.e., the severity level selected by the inspector). Optionally, the defect record object may also include remarks and generation time. Remarks are additional supplementary information entered by the inspector, and the generation time is the timestamp when the defect record object was created.

[0038] For example, the coordinates of the labeled area are rectangular frame coordinate data, including at least the horizontal coordinate (x), the vertical coordinate (y), the width (w), and the height (h). For instance, the coordinate data of a defect labeled area is {"x":108,"y":96,"w":280,"h":220}, indicating that the upper left corner of the rectangle is located at (108, 96) pixels in the inspection image, the width of the rectangle is 280 pixels, and the height is 220 pixels.

[0039] The inspection terminal sends the generated defect record object to the server via a wireless network. The server receives the defect record object, parses it, and extracts fields such as the defect image path, coordinates of the marked area, defect label, defect severity level, and on-site description. The server establishes an association between the defect record object and the inspection item identifier received in step 1, and stores it in the server's database.

[0040] By adopting the above method, the visual information (images and annotation boxes), semantic information (labels and severity levels), and textual description information (on-site instructions) of defects are uniformly organized into a structured defect record object, which enables the defect data to be efficiently stored, retrieved, and processed by the computer system, providing standardized data input for subsequent recommendations of specification clauses.

[0041] Step 3: Based on the defect label identifier, recall the set of candidate standard clauses from the preset mapping relationship between defect labels and standard clauses, and expand or filter the set of candidate standard clauses according to at least one of the following: on-site instructions, content of inspection items, construction location information, and process information.

[0042] For example, after receiving a defect record object, the server extracts the defect tag identifier. The server pre-stores a mapping table between defect tags and standard clauses. This mapping table records one or more standard clause identifiers corresponding to each defect tag identifier. This mapping relationship can be pre-configured by experts in the decoration and renovation engineering field, or it can be automatically constructed based on historical inspection data using machine learning methods. It is understood that standard clauses refer to specific provisions in national standards, industry standards, or enterprise standards related to the quality of decoration and renovation engineering, such as a specific provision in the "Standard for Acceptance of Construction Decoration and Renovation Engineering Quality" (GB 50210).

[0043] Specifically, the server queries the mapping table based on the defect tag identifier to obtain all the standard clause identifiers corresponding to that defect tag identifier. Then, it reads the complete clause content corresponding to these standard clause identifiers from the standard library to form the first candidate clause set. Each standard clause stored in the standard library includes at least the following fields: standard number (e.g., "GB50210-2018"), standard name (e.g., "Standard for Acceptance of Construction Decoration and Renovation Engineering Quality"), standard category (e.g., "National Standard"), version number (e.g., "2018 Edition"), clause content (e.g., "Facing bricks should be laid flat and firmly, and the joint width should be uniform"), effective date, and validity status (e.g., "Valid" or "Invalid").

[0044] Optionally, the defect label identifier corresponds to the leaf label node in the tree-shaped defect label system. The tree-shaped defect label system is a multi-level defect classification structure. In the description of this embodiment, the tree-shaped defect label system refers to a classification system that organizes various quality defects that may occur in decoration and renovation projects according to hierarchical relationships. For example, the first-level classification of this tree-shaped defect label system includes "surface installation defects," "painting project defects," "ceiling project defects," and "stone project defects," etc.; the second-level classification includes "wall surface finish," "wall and ceiling surface paint," "ceiling finishing," and "stone edging," etc.; the leaf labels include "hollow wall tiles," "wall tile color difference," "uneven joints," "wall cracks," "peeling," "contamination," "sagging," "uneven ceiling finishing," "uneven ceiling joints," "stone height difference," and "stone chipped edges," etc. The parent label node (e.g., "surface installation defects") is used to perform coarse-grained recall of candidate clauses; that is, when the mapping relationship corresponding to the leaf label cannot recall a sufficient number of candidate clauses, it can be elevated to the parent label for a broader recall. Leaf tag nodes are used to generate specific defect record objects.

[0045] An illustration of a tree-shaped defect labeling system is provided in this embodiment of the invention, as shown in Table 1.

[0046] Table 1. Example of a tree-shaped defect labeling system

[0047] Optionally, defect labels can be pre-associated with construction locations and procedures. For example, the label "hollow wall tiles" can be associated with construction locations such as "living room wall" and "bathroom wall," and with construction procedures such as "tiling," "grouting," and "leveling." This association can serve as a constraint for subsequent candidate clause filtering and contextual consistency judgment.

[0048] In one possible implementation, when the first candidate clause set obtained based on the defect label mapping relationship is empty, or the number of candidate clauses in the first candidate clause set is less than a preset threshold (e.g., less than 3), the server performs extended recall. Extended recall includes keyword-based retrieval, semantic retrieval, or a combination of both. Specifically, the server retrieves a second candidate clause set from the standard library based on at least one of the keywords in the site description, the content of the inspection items, and the construction location information. For example, the server performs word segmentation on the site description "the hollow area of ​​the wall tiles is about 0.12 square meters, and the gap width is uneven," extracts the keywords "hollow," "gap width," and "wall tiles," and then uses these keywords to perform a full-text search in the clause content field of the standard library, selecting the standard clauses containing these keywords as the second candidate clause set.

[0049] In another possible implementation, the server can also use semantic retrieval for extended recall. Specifically, the server converts the on-site instruction text into text vectors, and also converts the text content of each clause in the standard library into text vectors. Then, it calculates the semantic similarity (e.g., cosine similarity) between the on-site instruction text vector and the text vectors of each clause, and includes clauses with semantic similarity exceeding a preset threshold as members of the second candidate clause set.

[0050] The server merges and deduplicates the first and second candidate clause sets to obtain a merged set of candidate standard clauses. The specific method for merging and deduplication is as follows: using the unique identifier of the standard clause (e.g., standard number + clause number) as the key, the clauses in the two sets are merged, and only one copy of each identical clause is retained.

[0051] Optionally, the server filters out invalid clauses, duplicate versions, or clauses that are clearly inconsistent with the current construction site from the candidate specification clause set. Specifically, the server checks the validity status field of each candidate clause; if the value of this field is "invalid," the clause is removed from the candidate set. If multiple versions exist for the same standard number (e.g., GB50210-2010 and GB50210-2018), the version with a validity status of "valid" and a more recent effective date is prioritized. If a candidate clause contains clauses that are clearly irrelevant to the current construction site (e.g., the current construction site is "wall" but the clause only relates to "ceiling"), the clause is removed from the candidate set.

[0052] The above approach achieves a dual-path recall mechanism that combines precise recall based on defect tags with extended recall based on text retrieval. This ensures a strong correlation between the recall results and the defect type, while also improving the coverage and robustness of the recall through extended recall.

[0053] Step 4: Calculate the recommendation score for each candidate standard provision in the candidate standard provision set.

[0054] For example, after obtaining a set of candidate specification clauses, the server calculates a recommendation score for each candidate specification clause in the set. The recommendation score is determined based on at least two of the following five factors: the basic mapping weight between the defect label and the candidate specification clause; the text matching result between the on-site description in the defect record object and the content of the candidate specification clause; the consistency result between the content of the inspection item and the scope of application of the candidate specification clause; the validity status or version information of the candidate specification clause; and the priority adjustment value corresponding to the defect severity level.

[0055] In a preferred implementation, the recommendation score is determined based on all five factors mentioned above. The calculation of the recommendation score ensures that the final output recommendations are highly relevant to the defect labels, match the on-site instructions and inspection context, and also consider the validity of the standards and the severity of the defects, thereby improving the quality and usability of the recommendation results.

[0056] Specifically, the server obtains the basic mapping weight corresponding to each candidate specification clause. The basic mapping weight is derived from the association weight in the defect label-specification clause mapping table. This association weight reflects the degree of correlation between the defect label and the specification clause, and its value ranges from, for example, 0 to 100, with a higher value indicating a higher degree of correlation. The server divides this association weight by 100 to obtain the normalized basic mapping weight score. .

[0057] The server calculates a text matching score between the defect record object text and the candidate specification clause text. The defect record object text includes at least the content of the site description field, and optionally may also include the defect label name. The candidate specification clause text is the clause content field. The text matching score includes keyword matching degree, semantic similarity, or a combination of both. For example, the text matching score... Calculate as follows: ,in This refers to the keyword hit rate (e.g., the percentage of keywords in the on-site explanation that are matched within the text content). Semantic similarity (calculated based on the similarity between the text vector of the defect record object and the text vector of the candidate specification clause). For weighting coefficients, 0 ≤ ≤1. Semantic similarity can be calculated using cosine similarity, Euclidean distance, or other common vector similarity measures.

[0058] The server calculates a contextual consistency score between the inspection items, construction locations, or process information and the applicable scope of candidate specification clauses. For example, the contextual consistency score... Calculate as follows: ,in The inspection item matching result (reflecting the degree of matching between the content of the inspection items and the scope of application of the provisions). The result of matching the work process or part (reflecting the degree of matching between the current construction work process or part and the work process or part to which the clause applies). This represents the construction location matching result (reflecting the degree of matching between the current construction location and the construction location to which the clause applies). All the above matching results have been pre-normalized to the range of 0 to 1.

[0059] The server retrieves the validity status, version information, and effective time of the candidate specification clauses. For example, the correction value corresponding to the validity status... The value is 1.0 when the clause is valid and is the latest version, 0.7 when the clause is valid but not the latest version, and 0 when the clause is invalid or the clause is directly removed from the candidate specification clause set.

[0060] The server adjusts the priority of candidate specification clauses based on the severity of the defect. For example, the priority adjustment value corresponding to the severity of the defect is... The severity levels are mapped to corresponding numerical values: low corresponds to 0.2, medium corresponds to 0.5, high corresponds to 0.8, and critical corresponds to 1.0.

[0061] In a preferred implementation, the recommendation score is calculated according to the following formula:

[0062] in, This represents the recommendation score for the i-th candidate specification clause. This represents the basic mapping weight between the defect label and the i-th candidate specification clause. This represents the text matching score between the stated on-site description and the content of the i-th candidate specification clause. The score represents the contextual consistency between the content, construction location information, or process information of the inspection item and the scope of application of the i-th candidate specification clause. This represents the correction value corresponding to the validity status or version information of the i-th candidate specification clause. This indicates the priority adjustment value corresponding to the severity level of the defect. , , , , These are the weighting coefficients, and the sum of the five factors equals 1.

[0063] Optionally, the scores of the above items are normalized before entering the weighted calculation so that they fall into the same numerical range (e.g., between 0 and 1) to ensure that the scores of different dimensions are comparable when weighted summation.

[0064] By using the above multi-factor weighted scoring method, the calculation of the recommended score integrates prior knowledge of the defect label (mapping weight), actual on-site conditions (text matching), inspection context (consistency), the authority of the standard (effectiveness), and the urgency of the defect (severity level), making the scoring results more comprehensive, objective, and accurate.

[0065] Step 5: Sort the candidate standard provisions in the candidate standard provision set according to the recommendation score, output the recommended standard provisions to the inspection terminal, and output the standard number, standard name, provision content and recommendation basis of the recommended standard provisions for the inspection terminal to display.

[0066] For example, after calculating the recommendation score for each candidate specification clause in the candidate specification clause set, the server sorts the candidate specification clauses in the set according to the recommendation scores from highest to lowest. After sorting, the server selects a specified number of the top-ranked candidate specification clauses from the sorting results as recommended specification clauses, based on a preset output quantity (e.g., Top-3 or Top-5).

[0067] The server outputs recommended specification clauses to the inspection terminal. Specifically, the server encapsulates the recommended specification clauses, their corresponding standard numbers, standard names, clause content, and recommendation basis into a response data packet, which is then sent to the inspection terminal via the wireless network. Upon receiving this response data packet, the inspection terminal displays the standard number, standard name, clause content, and recommendation basis of the recommended specification clauses on its screen.

[0068] For example, the recommendation criteria should include at least two of the following items: the name or code of the defect label that was hit (e.g., "Hit defect label: hollow wall tiles"); the keywords, similar text fragments, or semantic recall sources that were hit (e.g., "Hit keywords: hollow, flat, joint width"); a description of consistency with the inspection items, construction locations, or procedures (e.g., "Consistent with the scenarios of 'wall tile flatness inspection' and 'living room wall'"); the validity status, version information, or effective time of the clause (e.g., "clause status: valid"); and the ranking position of the candidate clause in the current candidate set (e.g., "ranking position: 1st").

[0069] By simultaneously displaying the basis for the recommendations, the recommendations become more interpretable. Inspectors can clearly understand why each recommendation was made, facilitating on-site confirmation and subsequent review, and enhancing their trust in the recommendations.

[0070] In one possible implementation, when the candidate specification clause set is empty, the server performs a text retrieval recall based on the content of the site instructions and inspection items. That is, it skips the recall path based on defect tag mapping and directly uses keywords from the site instructions and inspection item content to perform a full-text search in the specification standard library, using the search results as the final candidate clause set.

[0071] In one possible implementation, when the recommended scores of all candidate specification clauses are lower than a preset threshold (e.g., 0.5), the server outputs a "manual review required" message and simultaneously displays at least one valid candidate specification clause with the highest score for inspection personnel to refer to and make a manual judgment.

[0072] In one possible implementation, when multiple versions of the same standard number exist, the server only retains candidate clauses with priority given to valid status and newer versions, avoiding recommending expired standard versions to inspection personnel.

[0073] In one possible implementation, when the on-site description text is too short (e.g., less than 3 Chinese characters), the server prioritizes the defect label, inspection items, and construction location for recommendations, reducing the weight of text matching score in the total score to ensure that the recommendation results do not become invalid due to insufficient on-site description information.

[0074] In one possible implementation, when the number of candidate entries is too large (e.g., more than 50), the server first performs a coarse screening based on the basic mapping weight or text matching results (e.g., only retaining the top 20 entries in terms of score), and then performs a final scoring and sorting on the coarsely screened candidate set to improve computational efficiency.

[0075] Based on the above technical solutions, this invention addresses the problem of difficulty in accumulating and reusing defect records in existing inspections by structurally associating inspection images, inspection context information, and defect record objects containing labeled coordinates, defect tags, and severity levels. This transforms scattered image and text descriptions into standardized data that can be efficiently stored, retrieved, and processed by computer systems, solving the problem of defect records being difficult to accumulate and reuse in existing inspections. Through a dual-path recall mechanism combining defect tag mapping-based recall with extended retrieval based on on-site descriptions and inspection item content, along with a multi-factor weighted scoring and ranking method, automatic recommendation of standard clauses is achieved. This reduces the workload of inspection personnel manually searching standard documents, improves clause retrieval efficiency, and ensures consistency in the clauses cited by different personnel and in different projects for the same type of defect. By synchronously outputting the standard number, standard name, clause content, and recommendation basis, the recommendation results are interpretable and facilitate on-site confirmation by inspection personnel.

[0076] In addition, the robustness of the solution in real inspection scenarios is improved through anomaly handling mechanisms such as empty recall downgrade, low confidence prompts, and version filtering.

[0077] The technical solutions provided in the embodiments of the present invention will be described and explained in detail below with reference to specific implementation scenarios.

[0078] Taking the scenario of hollow wall tiles as an example, the information entered by the inspectors is shown in Table 2.

[0079] Table 2. Example of Defect Record Input by Inspection Personnel

[0080] The system first retrieves the first set of candidate clauses from the mapping table based on the defect label "hollow tiles". Then, it supplements the second set of candidate clauses from the standard library by combining text matching results such as "hollow tiles", "uneven gap width", and "flatness of wall tile installation". After merging, deduplicating, and filtering invalid versions, three main candidate clauses are obtained. An example of their scoring calculation is shown in Table 3.

[0081] Table 3. Example of a rating table

[0082] Assuming that the weighting coefficient is used in this embodiment , , , , And the severity level medium corresponds to The candidate clause sorting process is as follows: Figure 2 As shown.

[0083] Therefore, the Top-3 recommendation results output by the system should include at least: 1. Candidate A: GB50210-2018-4.2.6 / Standard for Acceptance of Construction Decoration and Renovation Engineering Quality / The laying of facing bricks should be flat and firm, and the joint width should be uniform. There should be no obvious hollow, crooked or chipped edges.

[0084] 2. Summary of the clause corresponding to candidate B; 3. Summary of the clause corresponding to candidate C.

[0085] The recommendation criteria in this embodiment can be written as follows: 1. Defect tag hit: Hollow wall tiles; 2. Key words to target: hollow areas, flatness, seam width; 3. Contextual consistency: Consistent with the scenarios of "wall tile flatness inspection" and "living room wall"; 4. Clause status: Valid; 5. Sort position: 1st.

[0086] Taking the scenario of uneven ceiling finishing as an example, the information entered by the inspectors is shown in Table 4.

[0087] Table 4. Example of Defect Record Input by Inspection Personnel

[0088] The system first recalls candidate clauses based on the defect label "uneven ceiling finish" and the inspection context, and then performs secondary filtering based on information such as "height difference," "unstraight edge line," and "stone junction in the lobby." In this embodiment, the severity level "high" corresponds to... The scoring process is illustrated in Table 5.

[0089] Table 5. Example of a rating table

[0090] Therefore, the system prioritizes outputting GB50210-2018-6.3.4 / Standard for Acceptance of Construction and Decoration Engineering Quality / The installation of ceiling joists and panels should be firm, the surface flat, the joints straight, and there should be no looseness, arching, or significant height difference. This is the first recommended clause, and the following recommendation basis is provided simultaneously: 1. Defect tag: Uneven ceiling finish; 2. Key words: height difference, uneven finishing line, straight seam; 3. Contextual consistency: Consistent with the ceiling junction scenario involved in the "Lobby Stone Finishing Inspection"; 4. Clause status: Valid; 5. Sort position: 1st.

[0091] The embodiments of the present invention, through the above technical solutions, have at least the following beneficial effects: 1. This invention achieves structured data storage of defect annotation results by unifying inspection images, defect annotation areas, defect labels, and defect severity levels into the same defect record object.

[0092] 2. The embodiments of the present invention use tree-shaped defect labels and location-process mapping relationships to recall candidate clauses, thereby improving the consistency and relevance of the recommended results for similar defects.

[0093] 3. The embodiments of the present invention reduce the workload of inspection personnel in manually reviewing standard provisions through candidate provision recall, filtering, and sorting recommendation mechanisms, thereby improving the efficiency and stability of provision citation.

[0094] 4. By synchronously displaying the basis for the recommendation in the output results, the recommended clauses are interpretable, facilitating on-site confirmation and subsequent review.

[0095] 5. By introducing a multi-factor scoring calculation method, the candidate clause ranking process is made clearer, which helps to explain the logic of the recommendation results and improves the persuasiveness of the solution.

[0096] 6. The embodiments of the present invention improve the robustness of the solution in real inspection scenarios by adding empty recall downgrade, low confidence prompts and version filtering mechanisms.

[0097] 7. The recommended specifications output by the embodiments of the present invention can be reused by users in multiple scenarios such as rectification suggestions, quality descriptions, and report preparation, and have strong versatility.

[0098] On the other hand, embodiments of the present invention also provide a system for recommending standard clauses of decoration and renovation engineering based on defect annotation, the system including an inspection terminal and a server.

[0099] The inspection terminal includes an image acquisition module, a defect annotation module, and a result display module.

[0100] The image acquisition module is used to acquire inspection images, associate these images with inspection item identifiers and construction location information, and then send the images to the server. Specifically, the image acquisition module uses the camera of the inspection terminal to capture inspection images of the construction site, receives inspection item identifiers and construction location information input or selected by the inspection personnel through the interactive interface of the inspection application, and then encapsulates the inspection images and their associated information before sending them to the server.

[0101] The defect annotation module receives user selection and annotation operations for defect areas on inspection images, generates defect annotation area data, and combines this data with defect labels, defect severity levels, and on-site descriptions to generate defect record objects, which are then sent to the server. Specifically, the defect annotation module receives touch selection operations from inspectors on the image display interface of the inspection application, generating rectangular box coordinate data; it receives defect labels and defect severity levels selected by inspectors via drop-down menus or selection lists; it receives on-site descriptions entered by inspectors via text input boxes; and then, based on the above information, generates structured defect record objects and sends them to the server.

[0102] The results display module receives and displays the recommended specification clauses, their corresponding standard numbers, standard names, clause content, and recommendation basis returned by the server. Specifically, the results display module shows the recommended specification clauses in a list format on the inspection terminal screen. Each recommended clause displays its standard number, standard name, clause content, and recommendation basis, making it easy for inspection personnel to review and reference.

[0103] The server includes a candidate clause recall module, a scoring and sorting module, and a recommendation output module.

[0104] The candidate clause recall module receives inspection images and defect records. Based on defect tags, it recalls a set of candidate standard clauses from a pre-defined mapping between defect tags and standard clauses. It then expands or filters the set of candidate standard clauses based on at least one of the following: site instructions, inspection item content, construction location information, and process information. The specific operation of this module is consistent with the recall and filtering processes in steps 3 and 4 of the aforementioned method, and will not be repeated here.

[0105] The scoring and ranking module calculates a recommended score for each candidate specification clause in the candidate specification clause set. The recommended score is determined based on at least two of the following: the basic mapping weight between the defect label and the candidate specification clause; the text matching result between the on-site instructions and the content of the candidate specification clause; the consistency result between the content of the inspection items and the applicable scope of the candidate specification clause; the effective status or version information of the candidate specification clause; and the priority correction value corresponding to the defect severity level. The module then ranks the candidate specification clauses in the candidate specification clause set according to the recommended score. The specific operation of this module is consistent with the scoring and ranking process in step 5 of the aforementioned method, and will not be repeated here.

[0106] The recommended output module is used to return the sorted recommended specifications and their corresponding standard numbers, standard names, content, and basis for recommendation to the inspection terminal. The specific operation of this module is consistent with the output process in step 5 of the aforementioned method, and will not be repeated here.

[0107] Through the collaborative work of the inspection terminal and the server, the system provided by this invention realizes a complete automated process from defect labeling to standard clause recommendation, which greatly reduces the workload of inspection personnel manually checking standard clauses and improves the efficiency and consistency of clause citation.

[0108] It is understood that, in the description of the embodiments of this invention, the inspection terminal refers to a mobile intelligent device used for on-site quality inspection of decoration and renovation projects, such as a smartphone, tablet computer, or dedicated handheld inspection terminal. This terminal has image acquisition capabilities (e.g., a built-in camera), touch interaction capabilities (for receiving user selection and annotation operations on images), and wireless communication capabilities (for data interaction with a server). The server refers to a computer device or cloud computing platform deployed in the background, used to receive data uploaded by the inspection terminal, perform computationally intensive processing such as recalling and ranking standard provisions, and return recommendation results to the inspection terminal. The server can be a single physical server, a server cluster, or a cloud server instance.

[0109] For example, the system can also be an integrated device that has both server and terminal functions.

[0110] It should be noted that the module division in the embodiments of the present invention is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods. Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, exist as separate physical entities, or have two or more units integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0111] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0112] The computer program is stored in memory as a set of instructions. When the processor executes these instructions, it performs the steps of the aforementioned method, including receiving inspection images and defect record objects, recalling candidate specification clauses, calculating recommendation scores, and sorting and outputting recommendation results. This computer program product can be stored on computer-readable storage media such as optical discs, external hard drives, and USB flash drives, and can also be distributed and installed via network download.

[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for recommending clauses in decoration and renovation engineering specifications based on defect annotation, characterized in that, include: Receive inspection images sent by the inspection terminal, as well as the inspection item identifiers and construction location information associated with the inspection images; The inspection terminal receives a defect record object sent by the inspection terminal. The defect record object is generated by the inspection terminal based on the defect area selected and marked by the user on the inspection image. The defect record object includes at least the defect image path, the coordinates of the marked area, the defect label, the defect severity level, and the on-site description. Based on the defect label identifier, a set of candidate standard clauses is recalled from the preset mapping relationship between defect labels and standard clauses, and the set of candidate standard clauses is expanded or filtered based on at least one of the following: on-site instructions, content of inspection items, construction location information, and process information. A recommendation score is calculated for each candidate specification provision in the candidate specification provision set; The recommendation score is determined based on at least two of the following factors: the underlying mapping weight between the defect label and the candidate specification clause; The text matching results between the on-site descriptions in the defect record object and the content of the candidate standard clauses; the consistency results between the content of the inspection items and the scope of application of the candidate standard clauses; The validity status or version information of the candidate specification clauses; The priority correction value corresponding to the severity level of the defect; Based on the recommended scores, the candidate standard provisions in the candidate standard provision set are sorted, and the recommended standard provisions are output to the inspection terminal. The standard number, standard name, provision content, and recommendation basis of the recommended standard provisions are also output for the inspection terminal to display.

2. The method for recommending decorative and renovation engineering specification clauses based on defect annotation according to claim 1, characterized in that, The defect record object also includes remarks and generation time.

3. The method for recommending decorative and renovation engineering specification clauses based on defect annotation according to claim 1, characterized in that, The defect label identifier corresponds to the leaf label node in the tree-shaped defect label system, and the parent node in the tree-shaped defect label system is used for coarse-grained recall of candidate specification clauses; the defect label is pre-associated with the construction location and construction procedure.

4. The method for recommending decorative and renovation engineering specification clauses based on defect annotation according to claim 1, characterized in that, The step of recalling a set of candidate specification clauses from a preset mapping relationship between defect tags and specification clauses based on the defect tag identifier includes: Based on the defect label identifier, obtain the first set of candidate clauses from the defect label and specification clause mapping table; Based on at least one of the keywords in the site description, the contents of the inspection items, and the construction location information, a second set of candidate clauses is retrieved from the standard library. The first candidate clause set and the second candidate clause set are merged and deduplicated. When the first candidate text set is empty or the number is less than a preset threshold, an extended recall based on keyword retrieval, semantic retrieval, or a combination of both is performed.

5. The method for recommending decorative and renovation engineering specification clauses based on defect annotation according to claim 1, characterized in that, The text matching results include keyword matching degree and semantic similarity. The semantic similarity is calculated based on the similarity between the text vector of the defect record object and the text vector of the candidate specification clause.

6. The method for recommending decorative and renovation engineering specification clauses based on defect annotation according to claim 1, characterized in that, The recommendation score is calculated according to the following formula: in, This represents the recommendation score for the i-th candidate specification clause. This represents the basic mapping weight between the defect label and the i-th candidate specification clause. This represents the text matching score between the stated on-site description and the content of the i-th candidate specification clause. The score represents the contextual consistency between the content, construction location information, or process information of the inspection item and the scope of application of the i-th candidate specification clause. This represents the correction value corresponding to the validity status or version information of the i-th candidate specification clause. This indicates the priority adjustment value corresponding to the severity level of the defect. , , , , These are the weighting coefficients, and the sum of the five factors equals 1.

7. The method for recommending decorative and renovation engineering specification clauses based on defect annotation according to claim 6, characterized in that, The context consistency score is calculated by weighting the content matching results of the inspection items, the construction parts or procedures matching results, and the construction location matching results; the priority correction value corresponding to the severity level of the defect is mapped to the corresponding value in ascending order of severity; the correction value corresponding to the validity status or version information of the candidate specification clause is the first value when the clause is valid, the second value when the clause is valid but not a priority version, and the third value or the clause is removed from the candidate specification clause set when the clause is invalid.

8. The method for recommending decorative and renovation engineering specification clauses based on defect annotation according to claim 1, characterized in that, When the candidate specification clause set is empty, the text retrieval based on the content of the on-site instructions and the inspection items is downgraded; when the recommended scores of all candidate specification clauses are lower than the preset threshold, a manual review prompt is output and at least one candidate specification clause with the highest score is displayed; when there are multiple versions of the same standard number, the candidate specification clause with the most valid status and the latest version is retained.

9. A system for recommending standard clauses in decoration and renovation engineering based on defect annotation, characterized in that, Including inspection terminals and servers; The inspection terminal includes: The image acquisition module is used to acquire inspection images, associate the inspection images with inspection item identifiers and construction location information, and send the inspection images to the server. The defect annotation module is used to receive the user's box selection annotation operation on the inspection image, generate defect annotation area data, and generate a defect record object by combining the defect label, defect severity level and on-site description, and send the defect record object to the server. The results display module is used to receive and display the recommended specification clauses returned by the server, along with their corresponding standard numbers, standard names, clause content, and recommendation basis. The server includes: The candidate clause recall module is used to receive the inspection images and the defect record objects, recall a set of candidate standard clauses from the mapping relationship between the preset defect tags and standard clauses according to the defect tag identifier, and expand or filter the set of candidate standard clauses according to at least one of the following: on-site description, content of inspection items, construction location information and process information. The scoring and ranking module is used to calculate a recommendation score for each candidate specification clause in the candidate specification clause set. The recommendation score is determined based on at least two of the following: the basic mapping weight between the defect tag and the candidate specification clause; the text matching result between the on-site description and the content of the candidate specification clause; the consistency result between the content of the inspection item and the applicable scope of the candidate specification clause; the effective status or version information of the candidate specification clause; and the priority correction value corresponding to the defect severity level. The module then ranks each candidate specification clause in the candidate specification clause set according to the recommendation score. The recommended output module is used to return the sorted recommended standard provisions and their corresponding standard numbers, standard names, provisions, and recommendation basis to the inspection terminal.

Images