Semantic understanding-based completion environmental protection acceptance survey report auxiliary auditing system

Abstract

The invention relates to the technical field of natural language processing, in particular to a semantic understanding-based completion environmental protection acceptance survey report auxiliary auditing system, which comprises a structure recognition module, a semantic chain construction module, a semantic reconstruction module, an image-text coordinate fusion module and a spatial offset judgment module. In the method, through extracting subjects, terms and facility words in facility paragraphs and establishing a combination relationship, ordered labeling of semantic elements is realized, statement structure definition is enhanced, a semantic chain group is constructed and facility classification is associated, semantic consistency comparison is supported, semantic sequences and word frequency distribution in design and reply contents are compared, and description consistency is verified; facility coordinates in an image are extracted, character paragraphs are matched, image-text content anchoring, orientation, terrain and distance word extraction and conversion into space parameters to be compared with an image path are achieved, space consistency verification is completed, and the intelligent level of auditing on semantic comprehension, image-text fusion and space logic is overall improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural language processing, in particular to a completion environmental protection acceptance investigation report auxiliary auditing system based on semantic understanding. BACKGROUND

[0002] The technical field of natural language processing relates to the method of using computers to automatically analyze, understand and process natural language text. The core matters include semantic recognition, syntactic analysis, information extraction, text auditing and other contents. The overall technical field usually constructs language models, semantic rule sets, knowledge bases and other ways to structurally analyze the original text to identify the implied semantics in complex sentences and support intelligent auditing and content understanding applications. The traditional auditing of the completion environmental protection acceptance investigation report refers to the manual comparison of the report text content by the auditors according to relevant regulations and technical standards after the water and soil conservation facilities are completed, to check the consistency of the prevention and control measures and technical specifications, the logical rationality of the monitoring data, and the relevance of the information between different chapters. Usually, the human reading of regulations and clauses is combined with experience to determine whether the text expression conforms to the standard; the cross-checking of the specific contents such as water quality monitoring, soil loss calculation, and protection measure layout in the chapters is performed, and the compliance judgment, process tracking and version record management are completed manually.

[0003] The traditional auditing method relies on manual comparison of the contents of each chapter of the acceptance report, reading of the protection measures, water quality monitoring, soil loss calculation and other contents, and comparison with regulations and standards. The process needs to be compared item by item and relies heavily on the professional experience of personnel. Due to the non-standardization of the text structure and the complex nesting of semantic units in sentences, it is easy to cause confusion between the subject and object of the prevention and control measures, and the auditing personnel have difficulty in accurately positioning the combination relationship between the text subject and the term when judging whether the facilities meet the standard requirements, reducing the judgment efficiency. In the cross-chapter or cross-text comparison process, personnel often need to repeatedly review the design text and reply content, manually find the corresponding facility term for content correction, lack chain tracking ability, easily miss semantic inconsistency items, and reduce the checking integrity. The graphic information is not effectively linked, only relying on image page code prompts or manual searching for facilities in the image, there is a problem of mismatch between facility coordinates and text description, it is difficult to judge the consistency of the graphics and the text. The spatial description such as direction or distance lacks standardized expression, the auditing personnel need to subjectively interpret the text intention, which is difficult to directly verify with the position in the image, resulting in the lack of spatial description rationality verification. The above operation methods have the problems of high intensity of manual intervention, strong subjectivity of semantic interpretation, and weak matching ability of graphics and space, which restrict the overall auditing efficiency and accuracy. SUMMARY

[0004] To solve the technical problems in the prior art, the embodiment of the present application provides a completion environmental protection acceptance investigation report auxiliary auditing system based on semantic understanding.The technical solution is as follows: In one aspect, a completion environmental protection acceptance investigation report auxiliary auditing system based on semantic understanding is provided, which comprises: A structure identification module extracts a description paragraph of the acceptance report, processes a sentence by sentence, identifies the positions and labels of the subject, verb terms and facility names, records the character offset value and sequence index, establishes the combination relationship between the structure elements, and generates a paragraph sentence structure annotation set; A semantic chain construction module extracts the subject, terms and facility words based on the paragraph sentence structure annotation set, establishes a chain group structure in sequence, records the word form, word position index and paragraph number, associates the facility classification label, constructs a sequential chain structure, and generates a semantic element sequential chain group; A semantic reconstruction module matches the design file and the reply text paragraph based on the semantic element sequential chain group, compares the sequential offset, semantic intersection and word frequency distribution of the subject, terms and facility words, judges whether the chain group and the original text have semantic consistency, and generates a semantic traceability matching label set; A graphic-text coordinate fusion module extracts the facility area in the image and calculates the geometric center coordinates based on the semantic traceability matching label set, records the facility name, image page number and coordinate information, matches the report text paragraph number, constructs the facility position mapping, and generates a facility graphic-text cross page number and coordinate matching set.

[0005] As a further scheme of the present application, the paragraph sentence structure annotation set comprises a sentence structure label, a subject term offset, a facility name index and a structure combination relationship, the semantic element sequential chain group comprises a word form record, a word position index, a paragraph number and a facility classification label, the semantic traceability matching label set comprises a semantic consistency comparison result, a sequential offset information, a semantic intersection and a word frequency distribution, and the facility graphic-text cross page number and coordinate matching set comprises a facility name, an image page number and a facility coordinate.

[0006] As a further scheme of the present application, the structure identification module comprises: A sentence disassembly submodule obtains the position index of the punctuation in the sentence based on the extracted facility description content in the paragraph, judges whether the punctuation constitutes a semantic separation boundary, divides the original sentence content according to the semantic boundary, extracts the sub-sentence and the sequential position, and generates a sub-sentence sequential annotation information set; An element identification submodule calls the sub-sentence sequential annotation information set, identifies the position of the verb term in the sub-sentence, extracts the terms before and after the verb and judges the grammatical structure relationship, marks the character position and sequence number of the subject term, verb term and facility term, and generates a term position marking data group; The structure combination sub-module extracts the character order of the three types of terms in the original sentence according to the term position marking data set, judges whether the arrangement between the terms satisfies the structure combination condition, establishes the corresponding combination structure and index relationship between the terms, and obtains a sentence structure marking index set.

[0007] As a further scheme of the present application, the semantic chain construction module comprises: The element extraction sub-module obtains the sentence structure marking set, calls the character position and number information of the subject term, the verb term and the facility term, extracts the word form according to the term appearance order in the sentence, records the paragraph number where the term is located, establishes a term order set, and generates a term arrangement sequence value set; The index marking sub-module extracts the start position, end position and paragraph number of the term according to the term arrangement sequence value set, marks the index range of the term in the character sequence, calls the classification label corresponding to the facility term, establishes the corresponding relationship between the term index and the label, and generates a term position marking label set; The chain group construction sub-module calls the term position label data set, extracts the paragraph number and term order number, judges the connection relationship between the terms in the paragraph, constructs the term chain structure according to the term arrangement order, and obtains a semantic element order chain group.

[0008] As a further scheme of the present application, the semantic reconstruction module comprises: The paragraph matching sub-module obtains the semantic element order chain group, calls the number and term arrangement information of the design file paragraph and the reply text paragraph, extracts the corresponding paragraph position according to the term order number, establishes the corresponding structure of the design paragraph and the reply paragraph, and obtains a paragraph corresponding sequence value; The semantic comparison sub-module calls the word order number, word frequency and semantic intersection content of the terms, subject and facility words in the matching paragraph according to the paragraph corresponding sequence value, judges the word order offset relationship and semantic overlap range between the terms, establishes the paragraph term comparison structure, and generates a semantic position intersection coefficient; The consistency judgment sub-module calls the semantic position intersection coefficient, extracts the word frequency proportion relationship, semantic intersection quantity and word order offset position, judges the consistency state between the chain group paragraphs according to the semantic intersection range reference value, word frequency proportion reference value and word order offset reference range, and obtains a semantic trace matching label set.

[0009] As a further scheme of the present application, the image-text coordinate fusion module comprises: The image recognition sub-module obtains the semantic trace matching label set, extracts the boundary coordinate information of the facility region in the image, calls the image page number and the facility region corresponding pixel distribution, recognizes the coordinate point set in the boundary region, calculates the geometric center position of the facility region, and obtains a facility image center coordinate value; The position information recording sub-module extracts the facility name, image page code and corresponding coordinate position according to the facility image center coordinate value, calls the corresponding content of the facility label and image number, records the position information structure of the facility, establishes the position parameter set of the facility in the image, and obtains the facility image position information group; The paragraph mapping sub-module calls the facility image position information group, extracts the facility name and image page code information, identifies the number content of the paragraph where the facility first appears in the report text, establishes the numbering relationship between the facility and the text according to the corresponding rule of the facility name and the paragraph number, and generates the facility text-image cross page code and coordinate matching set.

[0010] As a further scheme of the present application, the system further comprises: The space offset judgment module extracts the direction, terrain and distance terms in the facility paragraph based on the facility text-image cross page code and coordinate matching set, uniformly converts them into angle values, landscape labels and distance units, establishes the space positioning parameters of the facility in the text description, compares the space path constituted by the facility coordinates in the image, and generates the text-image facility space description consistency mapping group; The text-image facility space description consistency mapping group comprises angle values, landscape labels, distance units and space path comparison results.

[0011] As a further scheme of the present application, the space offset judgment module comprises: The term extraction sub-module obtains the facility text-image cross page code and coordinate matching set, extracts the direction terms, terrain terms and distance terms in the facility paragraph, judges the semantic attributes of the terms, records the term content in the facility number and paragraph number, and generates the space description term group; The parameter conversion sub-module calls the space description term group, converts the direction terms into angle values, converts the terrain terms into landscape labels, and converts the distance terms into distance values in a uniform unit, establishes the space parameter content of the facility in the paragraph description, and obtains the text space positioning parameter set; The path comparison sub-module extracts the angle values, landscape labels and distance values of the facility according to the text space positioning parameter set, calls the facility coordinate information in the facility text-image cross page code and coordinate matching set, compares the direction, position and distance content in the paragraph space parameter and the coordinate path, establishes the comparison structure between the text and the image, and obtains the text-image facility space description consistency mapping group.

[0012] As a further scheme of the present application, the facility description refers to the text information associated with the facility, including the function, structure, position and use case descriptive information of the facility; The subject, verb term and facility name refer to a verb or verb phrase describing the function, operation or action of the facility, a term describing the operation, function description or state change of the device; The character offset value refers to the character distance of a word or term from the start of a text in a text; The sequential index refers to the sequential number of a word appearing in an identified sentence or paragraph; The semantic chain is a semantic link between words established by sorting the subject, verb and facility words in the facility description based on the syntactic structure and semantic relationship; The design file refers to an engineering design document associated with a facility; The batch text refers to an approval or review document given by an associated approval authority or authoritative institution; The semantic consistency refers to the comparison of the semantic differences between two texts in the grammatical structure, word meaning and context through semantic analysis technology; The image-text coordinate fusion refers to the fusion and alignment of the location information of a facility from the text and image by calculating the geometric center coordinates of the facility area in the image and combining the semantic traceability label to form an accurate facility location mapping.

[0013] As a further scheme of the present application, the spatial offset determination refers to converting the directions, terrain and distances in the description into standard spatial coordinates or angles, topographic labels based on natural language processing technology to form quantifiable spatial positioning data; The image-text facility spatial description consistency mapping is achieved by comparing the facility coordinates in the image with the spatial positioning data in the text description.

[0014] Compared with the prior art, the present application has the following advantages and positive effects: The present application realizes ordered labeling of semantic elements by extracting the subject, term and facility words in the facility paragraph and establishing a combination relationship, enhances the clarity of sentence structure, constructs semantic chain groups in order and associates facility classification, effectively supports semantic consistency comparison, compares the semantic order and word frequency distribution in the design and reply content, verifies the description consistency, improves the content traceability accuracy, extracts the facility coordinates in the image and matches the text paragraph, realizes image-text content anchoring, extracts the direction, terrain and distance words in the description, converts them into spatial parameters and image path comparison, completes the spatial consistency verification, and overall enhances the intelligentization and refinement of acceptance review in semantic understanding, image-text fusion and spatial logic. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0016] Figure 1 is a system flowchart of the present application; Figure 2 is a system block diagram of the present application; Figure 3 is a flowchart of a structure identification module of the present application; Figure 4 is a flowchart of a semantic chain construction module of the present application; Figure 5 is a flowchart of a semantic reconstruction module of the present application; Figure 6 is a flowchart of a text-image coordinate fusion module of the present application; Figure 7 is a flowchart of a spatial offset determination module of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the present application will be described below with reference to the drawings.

[0018] In the embodiments of the present application, the words such as "example", "for example" and the like are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0019] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. The words "of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0020] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0021] In order to make the technical problems, technical solutions and advantages to be solved by the present application more clear, the following will be described in detail with reference to the drawings and specific embodiments.

[0022] The embodiments of the present application provide a completion environmental protection acceptance investigation report auxiliary auditing system based on semantic understanding, as shown in the completion environmental protection acceptance investigation report auxiliary auditing system based on semantic understanding. Figures 1-2 The system includes: The structure recognition module extracts the facility description paragraph in the acceptance report, performs sentence processing on the sentence, identifies the positions and labels of the subject, verb terms and facility name, records the character offset value and sequence index, establishes the combination relationship between the structure elements, and generates the paragraph sentence structure annotation set; The facility description refers to the textual information associated with the facility, including the function, structure, location and use of the facility; The subject, verb terms and facility name refer to the verb or verb phrase describing the function, operation or action of the facility, device operation, function description, and state change terms; The character offset value refers to the character distance of a word or term from the start of the text in the text, which is used to identify the position of the word in the paragraph; The sequence index refers to the sequence number of the word in the recognized sentence or paragraph, which is used to represent the order of the element in the sentence; The semantic chain construction module extracts the subject, term and facility word based on the paragraph sentence structure annotation set, establishes the chain group structure in order, records the word form, word position index and paragraph number, associates the facility classification label, constructs the sequential chain structure, and generates the semantic element sequential chain group; The semantic chain is based on the syntactic structure and semantic relationship, and the semantic contact chain between words is established by sorting the subject, verb and facility word in the facility description; The semantic reconstruction module matches the design file and the reply text paragraph based on the semantic element sequential chain group, compares the sequence offset, semantic intersection and word frequency distribution of the subject, term and facility word, judges whether the chain group and the original text have semantic consistency, gets the comparison result, and generates the semantic trace matching label set; The design file refers to the engineering design document associated with the facility; The reply text refers to the approval or review document given by the associated approval authority or authoritative institution; Semantic consistency refers to comparing the semantic differences in grammar structure, word meaning and context between two texts through semantic analysis technology to ensure the matching degree of the semantics; The image-text coordinate fusion module extracts the facility area in the image and calculates the geometric center coordinates based on the semantic trace matching label set, records the facility name, image page number and coordinate information, matches the first appearance paragraph number in the report text, constructs the facility position mapping between the text and the image, and generates the facility image-text cross page number and coordinate matching set; Image-text coordinate fusion refers to calculating the geometric center coordinates of the facility area in the image, and combining the semantic trace label to fuse and align the position information of the facility from the text and the image, forming an accurate facility position mapping; The spatial offset determination module extracts the direction, terrain and distance terms in the facility paragraph based on the facility image-text cross page code and coordinate matching set, uniformly converts them into angle value, landform label and distance unit, constructs the spatial positioning parameters of the facility in the text description, compares with the spatial path composed of the facility coordinates in the image, and generates the image-text facility spatial description consistency mapping group; The spatial offset determination module extracts the direction, terrain and distance terms in the facility paragraph based on the facility image-text cross page code and coordinate matching set, uniformly converts them into angle value, landform label and distance unit, constructs the spatial positioning parameters of the facility in the text description, compares with the spatial path composed of the facility coordinates in the image, and generates the image-text facility spatial description consistency mapping group; The image-text facility spatial description consistency mapping is to compare the facility coordinates in the image with the spatial positioning data in the text description to ensure the consistency of the image and the text in the facility spatial description, and to generate the corresponding mapping relationship.

[0023] The paragraph sentence structure annotation set includes sentence structure label, subject term offset, facility name index, structure combination relationship, the semantic element order chain group includes word form record, word position index, paragraph number, facility classification label, the semantic source matching label set includes semantic consistency comparison result, order offset information, semantic intersection, word frequency distribution, the facility image-text cross page code and coordinate matching set includes facility name, image page code, facility coordinate, the image-text facility spatial description consistency mapping group includes angle value, landform label, distance unit, spatial path comparison result.

[0024] Specifically, as shown in Figure 2 , 3 , the structure recognition module includes: The sentence disassembly submodule obtains the position index of punctuation in the sentence based on the facility description content extracted in the paragraph, judges whether the punctuation constitutes a semantic separation boundary, divides the original sentence content according to the semantic boundary, extracts the sub-sentence and the order position, and generates the sub-sentence order annotation information set; The sentence disassembling submodule performs a character-level segmentation operation on the facility description class sentence. First, all punctuation character position indexes (such as commas, periods, semicolons, etc.) are extracted by traversing the entire sentence. Whether a semantic break is formed is determined according to the semantic structure calling rules in the window range before and after each punctuation. For example, in “the water supply pump is installed on the first floor underground, and is connected to the cooling tower inlet”, the comma position is identified and it is determined that the front segment is installation location information and the rear segment is connection relationship. After the semantic break rule is satisfied, the comma position is recorded as an effective segmentation point. Then, the original sentence is segmented according to the effective punctuation position, forming two clauses “the water supply pump is installed on the first floor underground” and “is connected to the cooling tower inlet”. Each clause is assigned a sequential number and the character start and end positions are recorded. At the same time, clause content, intra-sentence numbering, original text position, etc. are labeled to form a clause sequential labeling information set, which is used for subsequent structure analysis. For example, in another sentence “the water heater is arranged on the top of the corridor and is connected to the water storage tank through a copper pipe”, the system also identifies the comma as a semantic break point, extracts two clauses “the water heater is arranged on the top of the corridor” and “is connected to the water storage tank through a copper pipe”, records the sequence as 0 and 1, and combines them into a structured labeling set.

[0025] The element recognition submodule calls the clause sequential labeling information set, identifies the position of the verb term in the clause, extracts the terms before and after the verb and determines the grammatical structure relationship, labels the character positions and sequential numbers of the subject term, the verb term and the facility term, and generates a term position marking data group; After obtaining the clause sequential labeling information set, the element recognition submodule performs part-of-speech recognition and term positioning on each clause. First, the natural language processing tool is used to perform word segmentation and part-of-speech tagging on each clause. For example, the positions of the verbs “install”, “connect” and “set” are identified. Then, the nearest noun phrase is traced forward to determine the subject term, and the nearest noun phrase is traced backward to determine the facility term. In the sentence “the air outlet is connected to the air conditioner terminal device through the air pipe”, the system identifies “connect” as a verb, extracts “air outlet” as a subject term, and extracts “air conditioner terminal device” as a facility term. The start and end positions are recorded as 0 to 3 and 12 to 18, respectively. The clause number of the term is also labeled. Then, the term position marking data group is generated. For example, in another clause “the condensate pump is set on the west side of the basement”, the system identifies “set” as a verb, “condensate pump” as a subject term, and “basement west side” as a facility term. The start and end character positions in the original sentence and the clause number are recorded. The term position marking data group includes term category, term content, character index and clause identification, which is used for structure combination analysis.

[0026] The structure combination submodule extracts the character order of the three types of terms in the original sentence according to the term position marking data group, determines whether the arrangement between them satisfies the structure combination condition, establishes the corresponding combination structure and index relationship between the terms, and obtains a sentence structure labeling index set. The structural combination sub-module performs structural validity test and combination index labeling on the term position marking data set. According to the combination rule, it is judged whether the subject term is located on the left side of the verb term and whether the facility term is located on the right side of the verb term. The term triplets satisfying the condition are identified as valid structures and stored in the combination structure list. For example, in "the gas supply valve is arranged above the combustion device", the subject term "gas supply valve" is located in front of the action word "arrange", and the facility term "combustion device" is located behind the action word "arrange". The structure is in the order of subject-object-object, which meets the combination requirements, and the combination structure is labeled and recorded in the index set. If the sentence structure is "the cooling pipe is connected to the cooling pipe through the condenser", the subject term is missing or the order is not consistent, then it is judged as invalid structure and not recorded. The system also needs to record the combination start and end character positions, term order numbers, and clause positions of each valid structure index information, such as the two clauses in the sentence "the exhaust equipment is installed on the ceiling, and the connecting pipe leads to the fan". Both form valid term structure groups, respectively "exhaust equipment-install-ceiling" and "connecting pipe-lead-fan". The character index order meets the structure rule, so it is included in the sentence structure labeling index set, providing basic structure data for subsequent semantic graph construction.

[0027] Specifically, as shown in Figure 2 、 4 The semantic chain construction module includes: The element extraction sub-module obtains the paragraph sentence structure labeling set, calls the character position and number information of the subject term, verb term and facility term, extracts the word form according to the term appearance order in the sentence, records the paragraph number where the term is located, establishes the term order set, and generates the term arrangement sequence value set. The element extraction sub-module first receives the sentence structure labeling set, extracts the character position and order number of each group of subject term, verb term and facility term, sorts them according to the character index order during extraction, organizes and records the sorted term word form according to the actual arrangement order, reads the paragraph number information at the same time and synchronously stores it to form the term order set. When processing the sentence "the new fan unit is set on the roof platform and connected to the air outlet through the pipe", the system extracts 6 terms "new fan unit", "set", "roof platform", "pipe", "connect" and "air outlet" in turn, and the corresponding start position index is 0, 4, 8, 13, 15 and 18 respectively. The order number is reordered as 0 to 5, and these terms are uniformly bound with the paragraph number to form the term arrangement sequence value set. Each group of data includes the term original word, character start and end position, reordered order number and paragraph number. When processing multiple paragraphs such as "the drainage pump is set in the pit, and the outlet pipe is connected to the return tank", the system also extracts 6 term word forms and assigns order values, realizing unified management of term order structure under different sentence structures and forming a term order set covering the content of each paragraph.

[0028] The index labeling sub-module extracts the starting position, ending position and paragraph number of the term according to the term arrangement sequence value group, marks the index range of the term in the character sequence, calls the classification label corresponding to the facility term, establishes the corresponding relationship between the term index and the label, and generates a term position labeling label set; The index labeling sub-module calls the term arrangement sequence value group, extracts the character starting position, ending position and corresponding paragraph number of each term therein, and further reads the facility category label to which the term belongs, for example, the term "air supply pipe" has a starting position of 0, an ending position of 3, a paragraph number of 2, and a corresponding facility label of "pipe system category". The system binds the term index information and the label to form a labeling data structure, simultaneously identifies the action term "connect" starting at 4 and ending at 6, with a label of "behavior action category", and further identifies "air conditioner terminal equipment" starting at 7 and ending at 13, with a corresponding label of "terminal equipment category". The module records the starting and ending character positions, paragraph numbers and classification labels of all terms to form a term position labeling label set in a unified format. In another example, such as "cooling water pump is arranged on the east side of the equipment layer", the "cooling water pump" is a device term, "arranged" is a verb term, and "east side of the equipment layer" is a position term. The corresponding position index and paragraph number are labeled in sequence, combined with the term category label to construct a standardized data set. The position information of all terms is accurately labeled and forms a one-to-one corresponding structure with the classification label, so that subsequent operations can directly perform term matching and classification analysis through the index and the label.

[0029] The chain construction sub-module calls the term position label data group, extracts the paragraph number and term order number, judges the connection relationship between the terms in the paragraph, constructs the term chain structure according to the term arrangement order, and obtains a semantic element order chain group. The chain group building submodule reads the term position label data set, extracts the paragraph number and term order number of each term, and determines whether the arrangement relationship between the terms forms a continuous and effective structure according to the order. If the term numbers are continuous and the subject term, verb term and facility term have three roles, and all of them are in the same paragraph number, it is determined that it is an effective chain group structure, and the semantic element order chain group is added. In the example sentence "the return air pipe is connected to the condensing equipment, which is arranged inside the heat insulation layer", the system first identifies "return air pipe", "connect", "condensing equipment" as the first group of terms, the numbers are 0 to 2 in sequence, and the paragraph number is 0, forming the first chain group. Then, "set" is identified as a verb term and "heat insulation layer inside" is identified as a facility term. Since the subject term is missing, this group does not form an effective triple, which is excluded. In another scenario "the water supply pump is installed at the bottom of the pipe well, and the cooling water pipe is connected to the main pipe", the system identifies "water supply pump", "install", "pipe well bottom" as an effective triple term, the numbers are 0 to 2, the paragraph number is 1, forming chain group one. At the same time, "cooling water pipe", "connect", "main pipe" are identified as the second group of terms, the numbers are 3 to 5, and they belong to paragraph 1, forming chain group two. The two groups of term chains establish index structure and order mapping respectively, and finally output the semantic element order chain group. The structure contains paragraph number, term number order, term type label and character position index.

[0030] Specifically, as shown in Figure 2 、 5 The semantic reconstruction module includes: The paragraph matching submodule obtains the semantic element order chain group, calls the number and term arrangement information of the design file paragraph and the reply text paragraph, extracts the corresponding paragraph position according to the term order number, establishes the corresponding structure of the design paragraph and the reply paragraph, and obtains the paragraph corresponding sequence value. The paragraph matching sub-module reads the numbering information of the design file paragraphs and the reply text paragraphs in sequence after receiving the semantic element sequence chain group, and synchronously calls the term arrangement sequence data. In actual engineering, taking the building water supply and drainage drawing and the approval reply as an example, the system extracts the position number of the term in the character sequence from the design file paragraph, for example, in “the water supply pump is set in the underground first floor, and the water outlet is connected to the main pipe of the equipment layer”, the term number is 0 to 5, and the corresponding description in the reply text is “the pump position is in the specified area of the drawing, and the pipe connection direction is correct”, the term number is 2 to 6. The module first confirms that the number of terms in the two paragraphs is not more than 3 by comparing the number interval, then establishes the mapping of the term and the paragraph position according to the starting character index of the term in the original text, and finally forms a group of initial corresponding relationship between the design paragraph D1 and the reply paragraph R1. At the same time, the absolute sequence number of the term in the character sequence is extracted for cross-checking. The design term number 0 to 5 is compared with the reply term number 2 to 6, the matching situation is recorded item by item, and the paragraph numbering corresponding table and the term arrangement sequence comparison data are generated. In the scene of multiple paragraph comparison, if the design document contains continuous structure paragraphs D1 to D5, and the reply paragraphs are distributed in R1 to R4 respectively, the system will sequentially build multiple paragraph corresponding structures, and the output is the paragraph numbering mapping value set.

[0031] The semantic comparison sub-module calls the word sequence number, word frequency number and semantic intersection content of the terms, subjects and facility words in the matched paragraphs according to the paragraph corresponding sequence value, judges the word sequence offset relationship and semantic overlap range between the terms, establishes the paragraph term comparison structure, and generates the semantic position intersection coefficient. After the semantic comparison sub-module reads the paragraph corresponding sequence value set, it extracts the sequential number of all terms, subject structures and facility terms appearing in the corresponding paragraph. The frequency of each group of terms is counted in the comparison process. If "cooling water pump" appears 4 times in the design paragraph and 2 times in the reply paragraph, the word frequency ratio is 2 to 4, and the corresponding proportion is 50%. The module retains the smaller one as the semantic intersection quantity counting standard according to this proportion, extracts the common terms in the semantic intersection, such as "cooling", "setting", "equipment layer", etc., records the first occurrence position number in the paragraph, and calculates the word sequence offset value. For example, the term "setting" is numbered 1 in the design paragraph and 2 in the reply paragraph, so the offset value is 1. Record all term offsets and count the offset value range. In the example, the design paragraph term numbers are 0, 1, 2, 3, and 4, and the reply paragraph term numbers are 1, 2, 3, 4, and 5. The word sequence is shifted forward by 1 bit in total, so the offset range is 1. If the threshold is set to not more than 2, it is determined to be reasonable. The module establishes a term comparison structure based on the offset data, word frequency ratio and intersection term number, and stores each comparison value in the semantic position intersection data set. In engineering applications, for the term comparison process between mechanical and electrical installation drawings and approval opinion texts, the system uses word sequence numbers and semantic overlapping terms as the basis to judge the consistency of paragraph content item by item, providing comparison data support for subsequent consistency judgment.

[0032] The consistency judgment sub-module calls the semantic position intersection coefficient, extracts the word frequency proportion relationship, the semantic intersection quantity and the word sequence offset position, judges the consistency state between the chain group paragraphs according to the semantic intersection range reference value, the word frequency proportion reference value and the word sequence offset reference range, and obtains the semantic trace matching label set. The consistency judgment submodule reads the word frequency ratio, semantic intersection quantity and word sequence offset value recorded in the semantic position intersection data after obtaining the semantic position intersection data, and determines whether the paragraphs have a consistent relationship by comparing with the set reference value. For example, the word frequency ratio reference value is 40%, the semantic intersection word quantity reference value is 3, and the offset value is allowed to be in the range of ±2. When processing the sample "The air conditioner duct is set on the ceiling layer and connected to the end air outlet" and "The duct is set on the suspended ceiling layer and is consistent with the air outlet position", the system identifies the common terms as "duct", "set" and "air outlet", the quantity is 3, the word frequency is 3 times in the design paragraph and 2 times in the reply paragraph, the ratio is 2:3, i.e. 66%, and the offset value is calculated as an average offset of 1 number. All three indicators are within the set range, the system marks the paragraph matching as consistent, if the intersection of some group of paragraphs is only 2 terms or the word frequency ratio is less than 30%, it is determined as inconsistent. The final output is a semantic traceability matching label data set, including design paragraph number, reply paragraph number, intersection term quantity, word frequency ratio, word sequence offset value and final consistency judgment label. The label can be set as "consistent", "partially consistent" or "inconsistent". This structure can be widely used in design drawing review, engineering text tracing and other scenarios, and provides a structured basis for paragraph-level semantic review.

[0033] Specifically, as shown in Figure 2 、 6 the image-text coordinate fusion module includes: The image recognition submodule obtains a semantic traceability matching label set, extracts boundary coordinate information of the facility area in the image, calls the image page number and facility area corresponding pixel distribution, identifies the coordinate point set in the boundary area, calculates the geometric center position of the facility area, and obtains the facility image center coordinate value. The image recognition sub-module processes the facility area label information in sequence after obtaining the semantic traceability matching label set, extracts the page number and facility category of the image from the label by traversing each facility area marked in the image file, determines the specific boundary area of the facility in the image, and determines the geometric center point of the facility. In the engineering equipment layout drawing, for example, the "cooling pump A" marked in the equipment installation drawing is represented by a polygonal enclosure, and the boundary points include a plurality of two-dimensional coordinate points such as the position coordinates of the upper left corner, the upper right corner, the lower right corner and the lower left corner. The system extracts the horizontal and vertical coordinates of all boundary points in sequence, averages all horizontal coordinates to obtain the horizontal center position of the facility area, and averages all vertical coordinates to obtain the vertical center position, and finally determines the geometric center point of the facility. If the boundary points are regularly distributed, such as a rectangle, the center point coordinates can be directly obtained by averaging the coordinates, and if the boundary shape is irregular, such as a polygon or a curved boundary area, the center point can be determined by a polygon centroid calculation method or an area weighting method. When processing irregular shapes such as electrical wiring diagrams or equipment boundaries in computer room layouts, the center coordinates of each sub-region can be calculated by dividing the edge into several triangular regions, and the final center position can be obtained by weighted averaging. The processing result is output in the form of center point coordinates to represent the relative spatial position of the facility in the image.

[0034] The position information recording sub-module extracts the facility name, image page number and corresponding coordinate position according to the facility image center coordinate value, calls the corresponding content of the facility label and image number, records the position information structure of the facility, establishes the position parameter set of the facility in the image, and obtains the facility image position information group; After receiving the image center coordinate value of each facility area, the position information recording sub-module extracts the name, image page number and center position data of each facility in sequence, and one-to-one corresponds the facility name marked in the image to the actual image page by accessing the index relationship between the facility label and the image number. In the building drawing recognition scene, if the facility "valve V1" is located on the third page image with a center coordinate of 480 horizontally and 320 vertically, the information will be arranged as a data structure containing the name, page number and coordinate information. For facilities with the same name, such as "P1 pump" appearing in multiple images, the system distinguishes them by adding a label suffix such as "P1-A" and "P1-B". Each group of records will be supplemented with the coordinate points of the facility in the image to ensure the accuracy of spatial positioning. During the extraction process, if the image file embeds metadata such as facility identification number and image index number, the module will read and store the information in the position information simultaneously to form a complete position information parameter set for subsequent establishment of position association between text and image. In the facility information summary scene, this structure can be used to quickly locate the position of a device in the drawing, realize reverse image retrieval based on facility name, and finally arrange the position information group containing the names, image page numbers and spatial center coordinate values of all recognized facilities.

[0035] The paragraph mapping submodule calls the facility image position information set, extracts the facility name and image page information, identifies the paragraph number content of the first occurrence of the facility in the report text, constructs the numbering relationship between the facility and the image according to the corresponding rules of the facility name and the paragraph number, and generates the facility image-text cross page number and coordinate matching set; After obtaining the complete facility image position information set, the paragraph mapping submodule reads the facility name and corresponding image page information of each item in turn, searches for the paragraph number of the first occurrence of the corresponding facility in the text document, such as the report and the design specification, for example, the image of the facility named "cooling tower" is located on page 5, the system identifies the location paragraph number of the first occurrence of the name in the report as P12, establishes the corresponding structure of the facility name and the paragraph number, constructs the cross mapping relationship between the image and the text content, and in the case where the facility name appears repeatedly in multiple image pages, distinguishes them by combining the image page number, such as marking "cooling tower" on page 5 as "cooling tower-A" and marking "cooling tower" on page 6 as "cooling tower-B", further clarifying the relevance of the image and the paragraph, in the case where the facility name is expressed ambiguously, such as "P2 water pump" in the drawing and only "pump" or "main pump" in the text, the module uses semantic similarity matching and context analysis method to judge whether the facility in the text belongs to the target object, and then locates the paragraph number thereof, and finally forms a structured data set containing three related information of facility name, image page number and text paragraph number. The cross matching set can be used as basic data support for design document automatic proofreading, image-text matching analysis and other work.

[0036] Specifically, as shown in Figure 2 , 7 The spatial offset determination module includes: The word extraction submodule obtains the facility image-text cross page number and coordinate matching set, extracts the direction word, terrain word and distance word in the facility paragraph, determines the semantic attribute of the word, records the word content in the facility number and the paragraph number, and generates the spatial description word set. The word item extraction submodule, after receiving the facility map-text cross-page code and coordinate matching set, performs a structural semantic analysis operation on the paragraph where each facility first appears in the text. First, the paragraph text is syntactically analyzed to locate sentences with spatial description meanings, and words items describing the relative position of the facility are extracted, such as "north side", "due west", "left rear", etc. Then, topographic word items are extracted, such as "high ground", "low-lying", "terrace", etc. These word items are used to describe the topographic features of the facility. Subsequently, distance word items are identified by recognizing expressions containing numerical values and units, such as "about 10 m", "30 m from the main control room", "5 m outside the east wall", and the numerical value information and units are extracted and standardized. In the extraction process, the system processes each sentence in the paragraph using a morphological analysis tool and uses a pre-set direction word table, topographic word table, and distance expression template for matching and determination. For example, in the description "the main pump is located in the southeast direction of the main control building about 20 m away from the low-lying area", the system identifies "southeast direction" as a direction word item, "low-lying area" as a topographic word item, and "about 20 m" as a distance word item. It also corresponds to the facility number such as "PUMP001" and the text paragraph number such as "P15". The word item classification, number binding, and word item content recording are completed, and the final generated spatial description word item group will include the number of each facility, the paragraph number where it is located, the extracted direction, topography, and distance expression content, and will be organized in a unified format according to the structural rules.

[0037] The parameter conversion submodule calls the spatial description word item group, converts the direction word item into an angle value, converts the topographic word item into a topographic label, and converts the distance word item into a distance value in a unified unit to construct the spatial parameter content of the facility in the paragraph description and obtain the set of textual spatial positioning parameters. The parameter conversion submodule reads the generated spatial description term set, starts to perform parameter standardization operation, and converts the natural language expression of the orientation term into an angle value by establishing the corresponding relationship between the orientation and the angle, such as "east" corresponding to 90 degrees, "north" corresponding to 0 degrees, and "southwest" corresponding to 225 degrees. If there is a vague expression such as "north" or "near southeast" in the term, the median value in the preset fuzzy interval in the orientation term table is taken, such as "north" being set to 15 degrees and "near southeast" being set to 135 degrees. The conversion of the terrain term is realized by calling the topographic label dictionary, which maps "low-lying" to "LOWLAND", "mountain slope" to "SLOPE", and ensures the uniform expression of all terrain terms. The distance term needs to convert various units contained in the term into m. If it is km, multiply by 1000, such as "0.5km" converted to 500m. If it is cm, divide by 100, such as "300cm" converted to 3m. If the distance expression does not contain units such as "distance 20", the default unit is supplemented according to the context or the document. In the industrial scene, m is usually the default unit, so such cases are usually converted to 20m. In addition, it is also necessary to process terms containing modifying components such as "about" and "approximately". Such expressions are kept in the conversion process and are marked as approximate values, such as "about 50m" recorded as "50m (estimated)". All converted results form a unified structured spatial parameter record, including angle value, topographic label and distance value, and are bound to the corresponding facility number for subsequent comparative analysis.

[0038] The path comparison submodule extracts the angle value, topographic label and distance value of the facility according to the text spatial positioning parameter set, calls the facility coordinate information in the facility text and image cross page code and coordinate matching set, compares the direction, position and distance content in the paragraph spatial parameter and the coordinate path, establishes the comparison structure between the text and the image, and obtains the text facility spatial description consistency mapping group. The path contrast sub-module calls the constructed text space positioning parameter set and the text-image cross page code and coordinate matching set to perform space consistency evaluation processing. First, the facility image center point coordinates are extracted from the matching set, and a unified reference point coordinate is selected as a reference to calculate the direction of the facility center point relative to the reference point. By judging the difference between the horizontal and vertical coordinates, the relative orientation of the facility on the image is derived, and the orientation is compared with the angle value converted from the text space parameters. The angle deviation judgment range is set, such as the angle difference within 20 degrees is considered as consistent direction. Then, whether the terrain description of the facility is consistent is compared. The facility described as "LOWLAND" needs to fall within the image labeled lowland area. If the image area is not marked with a landform label, the cross judgment is performed according to the coordinate drop point and the layer annotation content, and then the distance value comparison is performed. The actual pixel distance between the reference point and the facility coordinate in the image needs to be converted into actual distance value. The pixel distance is converted into m units by reading the drawing scale information, such as a straight line in the figure represents a length of 20m, corresponding to 100px, the conversion ratio is 0.2m / px. If the pixel distance between the facility center point and the reference point is 150px, the actual distance is 30m. If the error does not exceed the set threshold such as ±2m, it is determined to be consistent. The comparison results of the three dimensions are summarized to form the final consistency evaluation results, and are recorded in the text-image facility space description consistency mapping group. Each record indicates the facility number, image page code, corresponding text paragraph number, and the flag indicating whether the three types of parameters are consistent, which is used for subsequent text-image synchronization verification and structured audit process.

[0039] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A semantic understanding-based auxiliary review system for environmental protection acceptance survey reports of completed projects, characterized in that: The system includes: The structure recognition module extracts the descriptive paragraphs from the acceptance report, processes the sentences into clauses, identifies the position and labels of subjects, verbs, terms, and facility names, records character offset values ​​and sequence indices, establishes the combination relationships between structural elements, and generates a set of paragraph sentence structure annotations. The semantic chain construction module extracts the subject, terminology and facility words based on the paragraph sentence structure annotation set, establishes a chain group structure in sequence, records word form, word position index and paragraph number, associates facility classification tags, constructs a sequential chain structure, and generates a semantic element sequential chain group. The semantic reconstruction module, based on the sequential chain of the semantic elements, matches the design document and the approval text paragraph, compares the order offset of the subject, terminology and facility words, semantic intersection and word frequency distribution, determines whether the chain group has semantic consistency with the original text, and generates a semantic source matching tag set. The image-text coordinate fusion module extracts facility areas from images and calculates their geometric center coordinates based on the semantic source matching tag set. It records facility names, image page numbers, and coordinate information, matches report text paragraph numbers, constructs facility location mappings, and generates a facility image-text cross-page number and coordinate matching set.

2. The semantic understanding-based auxiliary review system for environmental protection acceptance survey reports upon completion, as described in claim 1, is characterized in that: The paragraph sentence structure annotation set includes sentence structure tags, subject term offsets, facility name indexes, and structural combination relationships. The semantic element sequence chain group includes word form records, word position indexes, paragraph numbers, and facility classification tags. The semantic source matching tag set includes semantic consistency comparison results, sequence offset information, semantic intersection, and word frequency distribution. The facility image-text cross-page number and coordinate matching set includes facility names, image page numbers, and facility coordinates.

3. The semantic understanding-based auxiliary review system for environmental protection acceptance survey reports upon completion, as described in claim 1, is characterized in that: The structure recognition module includes: The sentence decomposition submodule extracts facility description content from paragraphs, obtains the position index of punctuation marks in sentences, determines whether punctuation marks constitute semantic separation boundaries, divides the original sentence content according to semantic boundaries, extracts clauses and their order positions, and generates a clause order annotation information set. The element recognition submodule calls the clause sequence annotation information set, identifies the position of verb terms in the clause, extracts terms before and after the verb and judges the grammatical structure relationship, annotates the character position and sequence number of subject terms, verb terms and facility terms, and generates term position tag data group; The structure combination submodule extracts the character order of the three types of terms in the original statement based on the term position tag data group, determines whether the arrangement between them meets the structure combination conditions, establishes the corresponding combination structure and index relationship between terms, and obtains the statement structure annotation index set.

4. The semantic understanding-based auxiliary review system for environmental protection acceptance survey reports upon completion, as described in claim 3, is characterized in that: The semantic chain construction module includes: The element extraction submodule obtains the paragraph sentence structure annotation set, calls the character position and number information of subject terms, verb terms and facility terms, extracts word forms according to the order of appearance of terms in the sentence, records the paragraph number where the term is located, establishes a term order set, and generates a term arrangement sequence value group. The index labeling submodule arranges the sequence value group according to the term, extracts the start position, end position and paragraph number of the term, marks the index range of the term in the character sequence, calls the classification label corresponding to the facility term, establishes the correspondence between the term index and the label, and generates a term position labeling label set. The chain group construction submodule calls the term position tag data group, extracts the paragraph number and term sequence number, determines the connection relationship between terms in the paragraph, constructs the term chain structure according to the term arrangement order, and obtains the semantic element sequence chain group.

5. The semantic understanding-based auxiliary review system for environmental protection acceptance survey reports upon completion, as described in claim 4, is characterized in that: The semantic reconstruction module includes: The paragraph matching submodule obtains the semantic element sequence chain group, calls the numbering and terminology arrangement information of the paragraphs in the design document and the approval text, extracts the corresponding paragraph position according to the terminology sequence number, establishes the corresponding structure of the design paragraph and the approval paragraph, and obtains the corresponding sequence value of the paragraph. The semantic comparison submodule calls the word order number, word frequency and semantic intersection content of terms, subjects and facilities in the matching paragraph according to the sequence value corresponding to the paragraph, determines the word order offset relationship and semantic overlap range between terms, establishes the paragraph term comparison structure and generates semantic position intersection coefficients. The consistency judgment submodule calls the semantic position intersection coefficient to extract the word frequency ratio relationship, the number of semantic intersections and the word order offset position. Based on the semantic intersection range benchmark value, the word frequency ratio benchmark value and the word order offset benchmark range, it judges the consistency status between the chain group segments and obtains the semantic source matching tag set.

6. The semantic understanding-based auxiliary review system for environmental protection acceptance survey reports for completed projects as described in claim 5, characterized in that: The image and text coordinate fusion module includes: The image recognition submodule obtains the semantic source matching tag set, extracts the boundary coordinate information of the facility area in the image, calls the image page number and the corresponding pixel distribution of the facility area, identifies the coordinate point set within the boundary area, calculates the geometric center position of the facility area, and obtains the center coordinate value of the facility image. The location information recording submodule extracts the facility name, image page number and corresponding coordinate position based on the center coordinate value of the facility image, calls the corresponding content of facility tag and image number, records the location information structure of the facility, establishes a set of location parameters of the facility in the image, and obtains the facility image location information group. The paragraph mapping submodule calls the facility image location information group, extracts the facility name and image page number information, identifies the number of the first paragraph in the report text where the facility appears, constructs the numbering relationship between the facility and the text based on the correspondence rules between the facility name and the paragraph number, and generates a set of facility image-text cross-page numbers and coordinate matching.

7. The semantic understanding-based auxiliary review system for environmental protection acceptance survey reports upon completion of projects, as described in claim 1, is characterized in that: The system also includes: The spatial offset determination module extracts the orientation, terrain and distance terms from the facility paragraphs based on the facility image and text cross page number and coordinate matching set, and converts them into angle values, landform labels and distance units. It constructs the spatial positioning parameters of the facility in the text description, compares them with the spatial path formed by the facility coordinates in the image, and generates a consistent mapping group of image and text facility spatial description. The consistency mapping group for spatial description of graphic facilities includes angle values, terrain labels, distance units, and spatial path comparison results.

8. The semantic understanding-based auxiliary review system for environmental protection acceptance survey reports upon completion of projects, as described in claim 7, is characterized in that: The spatial offset determination module includes: The term extraction submodule obtains the cross-page number and coordinate matching set of the facility image and text, extracts the location terms, terrain terms and distance terms in the facility paragraph, determines the semantic attributes corresponding to the terms, records the term content in the facility number and paragraph number, and generates a spatial description term group. The parameter conversion submodule calls the spatial description term group to convert directional terms into angle values, terrain terms into landform labels, and distance terms into distance values ​​in a uniform unit, thereby constructing the spatial parameter content of the facility in the paragraph description and obtaining the text spatial positioning parameter set; The path comparison submodule extracts the angle value, terrain label and distance value of the facility based on the text spatial positioning parameter set, calls the facility coordinate information in the facility image and text cross page number and coordinate matching set, compares the direction, position and distance content in the paragraph spatial parameters and coordinate path, establishes the comparison structure between the image and text, and obtains the image and text facility spatial description consistency mapping group.

9. The semantic understanding-based auxiliary review system for environmental protection acceptance survey reports upon completion of projects, as described in claim 1, is characterized in that: The facility description refers to the textual information associated with the facility, including descriptive information about the facility's function, structure, location, and usage. The subject, verb terms and facility names refer to verbs or verb phrases that describe the function, operation or action of the facility, and terms related to equipment operation, function description and state changes; The character offset value refers to the distance between a word or term and the character at the beginning of the text. The sequential index refers to the sequential numbering of words appearing in the identified sentence or paragraph; The semantic chain is based on syntactic structure and semantic relations. It establishes a semantic link between words by sorting the subject, verb and facility words in the facility description. The design documents refer to the engineering design documents associated with the facility; The approval text refers to the approval or review document issued by the relevant approval authority or authoritative body; The semantic consistency refers to comparing two texts using semantic analysis techniques to determine whether there are differences in grammatical structure, word meaning, and semantic context. The image-text coordinate fusion refers to calculating the geometric center coordinates of the facility area in the image and combining them with semantic source tags to fuse and align the facility's location information from the text and the image, forming an accurate facility location mapping.

10. The semantic understanding-based auxiliary review system for environmental protection acceptance survey reports upon completion, as described in claim 7, is characterized in that: The spatial offset determination refers to converting the location, terrain and distance in the description into standard spatial coordinates or angles and landform labels based on natural language processing technology, forming quantifiable spatial positioning data. The consistency mapping of spatial descriptions of facilities in images and texts is achieved by comparing the coordinates of facilities in the images with the spatial positioning data in the text descriptions.

Images