Building characterization carbon footprint estimation method and device based on artificial intelligence recognition and speculation

Abstract

The invention provides a building representation carbon footprint estimation method and device based on artificial intelligence recognition and speculation, and belongs to the technical field of carbon emission estimation. The method comprises the following steps: collecting building data to obtain a structure classification model data set and a material classification model data set; the method comprises the steps of calculating building feature values, establishing a building structure prediction model based on machine learning, replacing a feature extraction part of a YOLOV11 model with a SwinTransformerV2 or ConvNeXtV2 computer vision network based on the YOLOV11 model, and establishing a building material prediction model based on deep learning; collecting current situation data of a to-be-evaluated building, and obtaining a current situation characteristic value of the building through the model; constructing a carbon emission calculation model based on the building current situation characteristic values and the carbon emission factors; and calculating an estimated value of the building carbon emission. According to the method, the building material and carbon emission can be estimated more conveniently, efficiently and accurately.

Description

Technical Field

[0001] The present invention belongs to the technical field of carbon emission estimation, and in particular relates to a method and device for estimating the carbon footprint of building representation based on artificial intelligence recognition and inference. Background Art

[0002] With the acceleration of urbanization, the renovation of old buildings is closely related to carbon sink activities. Old buildings will become the main objects of carbon sink calculation and renovation. Through energy-saving renovation, increasing greening and other measures, carbon emissions can be effectively reduced and the carbon sink capacity of the city can be increased. However, the carbon sink calculation of old buildings is currently difficult. There are many old urban areas, and the relevant information and data are incomplete or difficult to obtain, which makes it difficult to accurately calculate carbon emissions.

[0003] The existing carbon emission measurement methods for old buildings mainly include:

[0004] 1. Geographic Information System (GIS) method: Use GIS technology to integrate urban construction, transportation, energy consumption and other data to conduct spatial analysis of carbon emissions. Disadvantages: Data acquisition and updating are difficult, and there may be missing or inaccurate information; in addition, the model is usually oversimplified and difficult to reflect the complex urban ecosystem.

[0005] 2. Life Cycle Assessment (LCA): Evaluate the carbon emissions of buildings and facilities during their life cycle, including the extraction, production, transportation, use and demolition of raw materials. Disadvantages: high implementation cost and long time; strong dependence on data, requiring a large amount of accurate material and energy data.

[0006] 3. Energy audit: Estimating the carbon emissions of buildings by auditing their energy consumption. Disadvantages: Only focusing on energy consumption data, not considering other sources of carbon emissions (such as transportation, waste disposal, etc.); and the audit process is cumbersome and difficult to apply on a large scale.

[0007] 4. Statistical methods: Use statistical data and models to estimate the city’s overall carbon emissions. Disadvantages: Usually based on macro-statistical data, it cannot accurately reflect the actual situation in a specific area; it may ignore local differences and the characteristics of individual buildings.

[0008] 5. Building energy consumption simulation: Use simulation software to predict building energy consumption and calculate carbon emissions. Disadvantages: The accuracy of the model depends on the accuracy of the input parameters, and the simulation results may be affected by simplified assumptions.

[0009] In summary, these methods have certain shortcomings in data acquisition, calculation accuracy and implementation feasibility, resulting in the carbon emission assessment of old urban areas being often not comprehensive and accurate. Summary of the invention

[0010] In response to the problems existing in the above-mentioned prior art, the present invention provides a method and device for estimating the carbon footprint of building representation based on artificial intelligence recognition and inference, combining computer vision and deep learning technology to carry out more convenient, efficient and accurate monitoring of building materials and carbon emissions.

[0011] Carbon footprint is an indicator used to measure the amount of carbon dioxide emissions caused directly or indirectly by an individual, organization, product or country within a certain period of time. The calculation of carbon footprint covers the emissions of the entire life cycle of a product or service from production, transportation, final use to disposal.

[0012] The present invention provides a method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and speculation, comprising the following steps:

[0013] Step S1: Collect data, collect structure classification model data, including building data and building point of interest POI data, and collect material classification model data, including street view images;

[0014] Step S2: pre-training data processing: for the structure classification model data, the buildings are labeled and matched with the points of interest to obtain the structure classification model data set; for the material classification model data, the data is segmented and labeled to obtain the material classification model data set;

[0015] Step S3: Establishing a machine learning-based building structure prediction model: extracting building features and calculating building feature values, evaluating and screening building features, determining a machine learning algorithm, establishing a machine learning-based building structure prediction model and training it, and obtaining a machine learning-based building structure prediction model that meets a preset accuracy;

[0016] Step S4: Establishing a deep learning-based building material prediction model: Performing data enhancement on the material classification model data set, based on the YOLOV11 model, using the SwinTransformerV2 or ConvNeXtV2 computer vision network to replace the feature extraction part of the YOLOV11 model, establishing a deep learning-based building material prediction model and training it, and obtaining a deep learning-based building material prediction model that meets the preset accuracy;

[0017] Step S5: Collect the status data of the building to be evaluated, obtain the building structure characteristic value and the building material characteristic value respectively through the building structure prediction model based on machine learning and the building material prediction model based on deep learning, and obtain the building status characteristic value;

[0018] Step S6: constructing a carbon emission calculation model based on building structure characteristics, building material characteristics and corresponding carbon emission factors;

[0019] Step S7: Obtain an estimated value of the building's carbon emissions based on the building's current characteristic values ​​and the carbon emissions calculation model.

[0020] Preferably, step S1 comprises the following steps:

[0021] Step S101: Obtain building data including building footprints, outlines, and attributes through a geographic information system, and collect building point of interest (POI) data through a network map.

[0022] Step S102: Collect street view images through real photos and online maps.

[0023] Preferably, step S2 comprises the following steps:

[0024] Step S201: Processing the structural classification model data, manually marking buildings with reference to building codes and technical standards, building-related publications, online satellite and street view maps, matching building data with point of interest (POI) data, and collating to obtain a complete structural classification data set in shp format;

[0025] Step S202: Use roboflow to segment and annotate the street view image to obtain an annotated material classification model dataset, and perform data enhancement on the annotated material classification model dataset.

[0026] Step S203: Divide the structure classification dataset and the street view dataset into a training set and a test set.

[0027] Preferably, the labeled material classification model data set is enhanced using techniques including random cropping, rotation, tilting, and noise blurring.

[0028] Preferably, step S3 comprises the following steps:

[0029] Step S301: extract building features; and calculate building feature values;

[0030] Step S302: Evaluate and screen building characteristics;

[0031] Step S303: Determine a machine learning algorithm and establish a building structure prediction model based on machine learning;

[0032] Step S304: training a building structure prediction model based on machine learning;

[0033] Step S305: Evaluate the prediction results of the machine learning-based building structure prediction model to obtain a machine learning-based building structure prediction model that meets a preset accuracy.

[0034] Preferably, evaluating and screening building characteristics includes:

[0035] Use the feature scoring function to score each feature;

[0036] Sort the features by their ratings from high to low.

[0037] Select features that rank in the top K of the preset positions.

[0038] Preferably, the building features include building geometry features, building shape features, and building point of interest (POI) features.

[0039] Preferably, the calculation of the building geometric characteristic value comprises the following steps:

[0040] Predefine the number of building floors and building height;

[0041] The height between floors is calculated using formula (1):

[0042]

[0043] Use the geometry calculation function in ArcGIS platform to calculate the outline area and perimeter of the building;

[0044] Simplify the building outline into the minimum bounding rectangle MBR surrounding the building, and calculate the length and width of the minimum bounding rectangle MBR through the minimum bounding geometry function of ArcGIS;

[0045] The length of the minimum bounding rectangle MBR is used to represent the true length of the building;

[0046] The width of the minimum bounding rectangle MBR is weighted and then used using formula (2) to represent the true width of the building outline:

[0047]

[0048] Calculate the ratio of the building's height to width and its length to width:

[0049] The building height to width ratio Rh2w is calculated using formula (3):

[0050]

[0051] The true length-to-width ratio of a building, Rl2w, is calculated using formula (4):

[0052]

[0053] Among them, H f is the height between floors, H b is the building height, N f is the number of floors of the building; W b is the true width of the building outline, A bis the outline area of ​​the building, A MBR is the area of ​​the minimum bounding rectangle MBR surrounding the building outline, L b is the actual length of the building, L MBR and W MBR are the length and width of the minimum bounding rectangle MBR respectively.

[0054] Preferably, the building shape characteristic value of the building is calculated using formula (5):

[0055]

[0056] Among them, SI is the shape irregularity of the building outline, and its value range is (0, 1.0]. The lower the value, the higher the irregularity.

[0057] Preferably, the calculation of the building point of interest POI feature value includes the following steps:

[0058] Count the total number of building points of interest POIs and the total number of building points of interest POI types for each building;

[0059] The building point of interest POI type is evaluated step by step using a decision tree until a leaf node is reached. The leaf node indicates the building point of interest POI type, and the corresponding Boolean value is obtained according to the building point of interest POI type.

[0060] Preferably, the determined machine learning algorithm is a gradient boosted decision tree.

[0061] Preferably, the building structure characteristic value obtained by the building structure prediction model based on machine learning includes the building structure type identified by the building structure prediction model based on machine learning, and the building structure volume Vmain body is calculated through street view image data, and the total volume of the building is determined according to the proportion of the building structure volume in different structure types, different building numbers of floors and heights.

[0062] Preferably, step S4 comprises the following steps:

[0063] Step S401: annotate the street view image to obtain annotated material classification model data set, and use random cropping, rotation, tilting, and noise blurring techniques to enhance the data;

[0064] Step S402: Based on the YOLOV11 model, the feature extraction part of the YOLOV11 model is replaced by the SwinTransformerV2 or ConvNeXtV2 computer vision network to establish a building material prediction model based on deep learning;

[0065] Step S403: Optimizing the hyperparameters of the deep learning-based building material prediction model to select the best hyperparameters;

[0066] Step S404: training the deep learning-based building material prediction model after the hyperparameter optimization;

[0067] Step S405: Evaluate the trained deep learning-based building material prediction model to obtain a deep learning-based building material prediction model that meets a preset accuracy.

[0068] Preferably, step S5 comprises the following steps:

[0069] Step S501: collecting current feature values ​​of building features through a geographic information system, including building geometric feature values, building shape feature values, and building point of interest POI feature values;

[0070] Step S502: Collecting street view images of the current status of buildings through online maps;

[0071] Step S503: The machine learning-based building structure prediction model and the deep learning-based building material prediction model respectively obtain the building structure characteristic value and the building material characteristic value, and obtain the building status characteristic value;

[0072] Step S504: sorting out building structure categories, building material proportions, building geometric feature values, building shape feature values, and building point of interest POI feature values.

[0073] Preferably, step S6 comprises the following steps:

[0074] Step S601: Collect existing data to obtain carbon emission factors C of different structures 主体 and carbon emission factors C of different materials 维护 ;

[0075] Step S602: Obtaining the area and volume of the building through the building geometric feature values;

[0076] Step S603: Construct a building structure carbon emission model and a building material carbon emission model. According to the area and volume of the building, combined with the carbon emission coefficient corresponding to its structure type and material, calculate the carbon emission of the main structure and the surrounding structure of the building. The carbon emission calculation formula is as follows:

[0077]

[0078] in:

[0079] V 主体i : The volume of the i-th structure (m 3 );

[0080] C 主体i : Carbon emission standard value of the i-th structure (kgCO2 / m 3 );

[0081] α i : Carbon emission coefficient of the i-th structure, used to adjust the impact of different structures;

[0082] A 围护j : Surface area of ​​the jth material (m 2 );

[0083] C 围护j : Carbon emission standard value of the jth material (kgCO 2 / m 2 );

[0084] β j : Carbon emission coefficient of the jth material, used to consider the characteristics of different materials;

[0085] E 其他 :Carbon emissions from other construction processes (kgCO 2 );

[0086] E 总 :Total carbon emissions (kgCO 2 );

[0087] Step S604: Evaluate the building structure carbon emission model and the building material carbon emission model;

[0088] Step S605: Obtain the final carbon emission calculation model.

[0089] The present invention provides a device for estimating the carbon footprint of building representation based on artificial intelligence recognition and inference, comprising a processor, wherein the processor executes the steps of the method for estimating the carbon footprint of building representation based on artificial intelligence recognition and inference.

[0090] Compared with the prior art, the present invention has at least the following beneficial effects:

[0091] 1. Efficient use of limited data

[0092] In view of the lack of data in old urban areas, this paper uses computer vision technology and deep learning models to directly extract key information such as building materials and structural characteristics by analyzing publicly available street view images and a small amount of existing building data. This method reduces the reliance on large-scale data, making it possible to accurately identify the characteristics of buildings and conduct high-quality carbon emission assessments even under limited data conditions.

[0093] 2. Low-cost data collection and processing

[0094] Unlike the traditional high-cost method that relies on comprehensive data collection and on-site sampling, this invention uses automated visual recognition technology to obtain building information with only public data sources and a small amount of manual annotation, thereby significantly reducing the cost of data collection and processing. This low-cost data processing method is suitable for situations with limited resources, making large-scale application of carbon emission assessment in old urban areas possible.

[0095] 3. Efficient calculation process

[0096] The present invention introduces a deep learning algorithm to efficiently process building information. Compared with traditional methods such as GIS analysis or life cycle assessment, the present invention can evaluate the carbon emissions of buildings in a short time, reducing a lot of manual intervention and data processing. This efficient assessment process greatly improves the efficiency of carbon footprint calculation and meets the needs of large-scale and rapid monitoring.

[0097] Through a low-cost and efficient solution under limited data conditions, the present invention provides a building carbon emission assessment method with strong adaptability and high resource utilization, providing important support for carbon emission monitoring and management in old urban areas.

[0098] 4. Accurate calculation

[0099] The present invention takes into account the structural type and material characteristics of the building, and innovatively provides a method for calculating the actual length and width of the building, as well as the irregularity of the building. The resulting building characteristics are more accurate, thereby making the calculated carbon emissions more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] Figure 1 It is a flowchart of a method for estimating carbon footprint of building representation based on artificial intelligence recognition and speculation according to an embodiment of the present invention.

[0101] Figure 2 It is a technical roadmap of a method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and speculation in an embodiment of the present invention.

[0102] Figure 3 It is a technical roadmap for constructing a building structure prediction model in an embodiment of the present invention to obtain a building structure prediction model that meets a preset accuracy.

[0103] Figure 4 It is a technical roadmap for constructing a building material prediction model in an embodiment of the present invention and obtaining a building material prediction model that meets a preset accuracy.

[0104] Figure 5 It is a technical roadmap for calculating carbon emissions according to the current characteristics of a building by using a building structure prediction model and a building material prediction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0105] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0106] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0107] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0108] In order to better understand the purpose, structure and function of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings.

[0109] The present invention provides a method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and speculation, comprising the following steps:

[0110] Step S1: Collect data, collect structure classification model data, including building data and building point of interest POI data, and collect material classification model data, including street view images;

[0111] Step S2: pre-training data processing: for the structure classification model data, the buildings are labeled and matched with the points of interest to obtain the structure classification model data set; for the material classification model data, the data is segmented and labeled to obtain the material classification model data set;

[0112] Step S3: Establishing a machine learning-based building structure prediction model: extracting building features and calculating building feature values, evaluating and screening building features, determining a machine learning algorithm, establishing a machine learning-based building structure prediction model and training it, and obtaining a machine learning-based building structure prediction model that meets a preset accuracy;

[0113] Step S4: Establishing a deep learning-based building material prediction model: Performing data enhancement on the material classification model data set, based on the YOLOV11 model, using the SwinTransformerV2 or ConvNeXtV2 computer vision network to replace the feature extraction part of the YOLOV11 model, establishing a deep learning-based building material prediction model and training it, and obtaining a deep learning-based building material prediction model that meets the preset accuracy;

[0114] Step S5: Collect the status data of the building to be evaluated, obtain the building structure characteristic value and the building material characteristic value respectively through the building structure prediction model based on machine learning and the building material prediction model based on deep learning, and obtain the building status characteristic value;

[0115] Step S6: constructing a carbon emission calculation model based on building structure characteristics, building material characteristics and corresponding carbon emission factors;

[0116] Step S7: Obtain an estimated value of the building's carbon emissions based on the building's current characteristic values ​​and the carbon emissions calculation model.

[0117] According to a specific embodiment of the present invention, step S1 comprises the following steps:

[0118] Step S101: Obtain building data including building footprints, outlines, and attributes through a geographic information system, and collect building point of interest (POI) data through a network map.

[0119] Step S102: Collect street view images through real photos and online maps.

[0120] According to a specific embodiment of the present invention, step S2 comprises the following steps:

[0121] Step S201: Processing the structural classification model data, manually marking buildings with reference to building codes and technical standards, building-related publications, online satellite and street view maps, matching building data with point of interest (POI) data, and collating to obtain a complete structural classification data set in shp format;

[0122] Step S202: Use roboflow to segment and annotate the street view image to obtain an annotated material classification model dataset, and perform data enhancement on the annotated material classification model dataset.

[0123] Step S203: Divide the structure classification dataset and the street view dataset into a training set and a test set.

[0124] According to a specific embodiment of the present invention, the labeled material classification model data set is enhanced using techniques including random cropping, rotation, tilting, and noise blurring.

[0125] According to a specific embodiment of the present invention, step S3 comprises the following steps:

[0126] Step S301: extract building features; and calculate building feature values;

[0127] Step S302: Evaluate and screen building characteristics;

[0128] Step S303: Determine a machine learning algorithm and establish a building structure prediction model based on machine learning;

[0129] Step S304: training a building structure prediction model based on machine learning;

[0130] Step S305: Evaluate the prediction results of the machine learning-based building structure prediction model to obtain a machine learning-based building structure prediction model that meets a preset accuracy.

[0131] According to a specific embodiment of the present invention, evaluating and screening building characteristics includes:

[0132] Use the feature scoring function to score each feature;

[0133] Sort the features by their ratings from high to low.

[0134] Select features that rank in the top K of the preset positions.

[0135] According to a specific embodiment of the present invention, the building features include building geometry features, building shape features, and building point of interest (POI) features.

[0136] According to a specific embodiment of the present invention, the calculation of the building geometric characteristic value includes the following steps:

[0137] Predefine the number of building floors and building height;

[0138] The height between floors is calculated using formula (1):

[0139]

[0140] Use the geometry calculation function in ArcGIS platform to calculate the outline area and perimeter of the building;

[0141] Simplify the building outline into the minimum bounding rectangle MBR surrounding the building, and calculate the length and width of the minimum bounding rectangle MBR through the minimum bounding geometry function of ArcGIS;

[0142] The length of the minimum bounding rectangle MBR is used to represent the true length of the building;

[0143] The width of the minimum bounding rectangle MBR is weighted and then used using formula (2) to represent the true width of the building outline:

[0144]

[0145] Calculate the ratio of the building's height to width and its length to width:

[0146] The building height to width ratio Rh2w is calculated using formula (3):

[0147]

[0148] The true length-to-width ratio of a building, Rl2w, is calculated using formula (4):

[0149]

[0150] Among them, H f is the height between floors, H b is the building height, N f is the number of floors of the building; W b is the true width of the building outline, A b is the outline area of ​​the building, A MBR is the area of ​​the minimum bounding rectangle MBR surrounding the building outline, L b is the actual length of the building, L MBR and W MBR are the length and width of the minimum bounding rectangle MBR respectively.

[0151] According to a specific embodiment of the present invention, the building shape characteristic value of the building is calculated using formula (5):

[0152]

[0153] Among them, SI is the shape irregularity of the building outline, and its value range is (0, 1.0]. The lower the value, the higher the irregularity.

[0154] According to a specific embodiment of the present invention, the calculation of the building point of interest POI feature value includes the following steps:

[0155] Count the total number of building points of interest POIs and the total number of building points of interest POI types for each building;

[0156] The building point of interest POI type is evaluated step by step using a decision tree until a leaf node is reached. The leaf node indicates the building point of interest POI type, and the corresponding Boolean value is obtained according to the building point of interest POI type.

[0157] According to a specific embodiment of the present invention, the determined machine learning algorithm is a gradient boosting decision tree.

[0158] According to a specific embodiment of the present invention, the building structure characteristic value obtained by the building structure prediction model based on machine learning includes the building structure type identified by the building structure prediction model based on machine learning, and the building structure volume Vmain body is calculated through street view image data, and the total volume of the building is determined according to the proportion of the building structure volume in different structure types, different building floors and heights.

[0159] According to a specific embodiment of the present invention, step S4 comprises the following steps:

[0160] Step S401: annotate the street view image to obtain annotated material classification model data set, and use random cropping, rotation, tilting, and noise blurring techniques to enhance the data;

[0161] Step S402: Based on the YOLOV11 model, the feature extraction part of the YOLOV11 model is replaced by the SwinTransformerV2 or ConvNeXtV2 computer vision network to establish a building material prediction model based on deep learning;

[0162] Step S403: Optimizing the hyperparameters of the deep learning-based building material prediction model to select the best hyperparameters;

[0163] Step S404: training the deep learning-based building material prediction model after the hyperparameter optimization;

[0164] Step S405: Evaluate the trained deep learning-based building material prediction model to obtain a deep learning-based building material prediction model that meets a preset accuracy.

[0165] According to a specific embodiment of the present invention, step S5 comprises the following steps:

[0166] Step S501: collecting current feature values ​​of building features through a geographic information system, including building geometric feature values, building shape feature values, and building point of interest POI feature values;

[0167] Step S502: Collecting street view images of the current status of buildings through online maps;

[0168] Step S503: The machine learning-based building structure prediction model and the deep learning-based building material prediction model respectively obtain the building structure characteristic value and the building material characteristic value, and obtain the building status characteristic value;

[0169] Step S504: sorting out building structure categories, building material proportions, building geometric feature values, building shape feature values, and building point of interest POI feature values.

[0170] According to a specific embodiment of the present invention, step S6 comprises the following steps:

[0171] Step S601: Collect existing data to obtain carbon emission factors C of different structures 主体 and carbon emission factors C of different materials 维护 ;

[0172] Step S602: Obtaining the area and volume of the building through the building geometric feature values;

[0173] Step S603: Construct a building structure carbon emission model and a building material carbon emission model. According to the area and volume of the building, combined with the carbon emission coefficient corresponding to its structure type and material, calculate the carbon emission of the main structure and the surrounding structure of the building. The carbon emission calculation formula is as follows:

[0174]

[0175] in:

[0176] V 主体i : The volume of the i-th structure (m 3 );

[0177] C 主体i : Carbon emission standard value of the i-th structure (kgCO 2 / m 3 );

[0178] α i : Carbon emission coefficient of the i-th structure, used to adjust the impact of different structures;

[0179] A 围护j : Surface area of ​​the jth material (m 2 );

[0180] C 围护j : Carbon emission standard value of the jth material (kgCO 2 / m 2 );

[0181] β j : Carbon emission coefficient of the jth material, used to consider the characteristics of different materials;

[0182] E 其他 :Carbon emissions from other construction processes (kgCO 2 );

[0183] E 总:Total carbon emissions (kgCO 2 );

[0184] Step S604: Evaluate the building structure carbon emission model and the building material carbon emission model;

[0185] Step S605: Obtain the final carbon emission calculation model.

[0186] The present invention provides a device for estimating the carbon footprint of building representation based on artificial intelligence recognition and inference, comprising a processor, wherein the processor executes the steps of the method for estimating the carbon footprint of building representation based on artificial intelligence recognition and inference.

[0187] Example 1

[0188] like Figure 1 The present invention provides a method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and speculation, comprising the following steps:

[0189] Step S1: Collect data, collect structure classification model data, including building data and building point of interest POI data, and collect material classification model data, including street view images;

[0190] Step S2: pre-training data processing: for the structure classification model data, the buildings are labeled and matched with the points of interest to obtain the structure classification model data set; for the material classification model data, the data is segmented and labeled to obtain the material classification model data set;

[0191] Step S3: Establishing a machine learning-based building structure prediction model: extracting building features and calculating building feature values, evaluating and screening building features, determining a machine learning algorithm, establishing a machine learning-based building structure prediction model and training it, and obtaining a machine learning-based building structure prediction model that meets a preset accuracy;

[0192] Step S4: Establishing a deep learning-based building material prediction model: Performing data enhancement on the material classification model data set, based on the YOLOV11 model, using the SwinTransformerV2 or ConvNeXtV2 computer vision network to replace the feature extraction part of the YOLOV11 model, establishing a deep learning-based building material prediction model and training it, and obtaining a deep learning-based building material prediction model that meets the preset accuracy;

[0193] Step S5: Collect the status data of the building to be evaluated, obtain the building structure characteristic value and the building material characteristic value respectively through the building structure prediction model based on machine learning and the building material prediction model based on deep learning, and obtain the building status characteristic value;

[0194] Step S6: constructing a carbon emission calculation model based on building structure characteristics, building material characteristics and corresponding carbon emission factors;

[0195] Step S7: Obtain an estimated value of the building's carbon emissions based on the building's current characteristic values ​​and the carbon emissions calculation model.

[0196] Example 2

[0197] refer to Figure 2-5 The present invention provides a method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and speculation, comprising the following steps:

[0198] 1. Construct a building structure prediction model and a building material prediction model to obtain a building structure prediction model and a building material prediction model that meet the preset accuracy:

[0199] Step S1: Collect data, collect structure classification model data, including building data and building point of interest POI data, and collect material classification model data, including street view images;

[0200] Wherein, step S1 comprises the following steps:

[0201] Step S101: Obtain building data including building footprints, outlines, and attributes through a geographic information system, and collect building point of interest (POI) data through a network map.

[0202] Step S102: Collect street view images through real photos and online maps.

[0203] Step S2: pre-training data processing: for the structure classification model data, the buildings are labeled and matched with the points of interest to obtain the structure classification model data set; for the material classification model data, the data is segmented and labeled to obtain the material classification model data set;

[0204] Wherein, step S2 comprises the following steps:

[0205] Step S201: Process the structure classification model data, refer to building codes and technical standards, building-related publications, online satellite and street view maps to manually mark buildings, match building data with point of interest POI data, and organize to obtain a complete shp format structure classification data set (data set 2);

[0206] Define structural tags and classify the structural classification model data into five common building structure types - masonry structure, concrete frame, concrete frame-shear wall, concrete shear wall and steel structure - as structural tags. POI data matching the selected buildings were captured from Amap and represented in a four-tuple format: {POI ID, x-coordinate, y-coordinate, POI type code}. Among them, "POI ID" is the unique identifier of each POI; "x-coordinate" and "y-coordinate" are the latitude and longitude coordinates of the POI in the Amap coordinate system. These coordinates were then converted to the same coordinate system as the .shp file, the WGS1984 Mercator World Coordinate System, in order to match the POI with the building.

[0207] To ensure the correctness of the annotations, the following contents were referenced:

[0208] Building codes and technical standards may contain distinguishing features. For example, the Code for Seismic Design of Buildings GB50011-2010 stipulates that masonry residential buildings in Beijing must withstand an earthquake intensity of 8 degrees and a peak acceleration of 0.2g with a height-to-width ratio of no more than 2.0, no more than 6 floors, and a height of no more than 18 meters.

[0209] Related publications, such as journal articles, construction documents, and technical reports, may shed light on the structural type of the building.

[0210] Online satellite and street view maps (such as Amap

[20] , the Chinese version of Google Maps) are used for visual identification of building structures.

[0211] Based on the above specifications and publications, the present invention proposes three assumptions to avoid confusion and resolve labeling conflicts: first, all masonry buildings have less than six floors; second, the concrete frame structure of public buildings has less than 10 floors, while the concrete frame-shear wall structure has 10 floors or more; finally, the concrete shear wall structure of residential buildings has 20 floors or more, while the concrete frame-shear wall structure has 7 to 19 floors, and the concrete frame structure has less than 7 floors.

[0212] Step S202: Use roboflow to segment and annotate the street view image to obtain an annotated material classification model dataset. Perform data enhancement on the annotated material classification model dataset (dataset 1) using techniques including random cropping, rotation, tilting, and noise blurring.

[0213] Step S203: Divide the structure classification dataset and the street view dataset into a training set and a test set.

[0214] Step S3: Establishing a building structure prediction model based on machine learning: extracting building features and calculating building feature values, evaluating and screening building features, determining that the machine learning algorithm is a gradient boosting decision tree, establishing a building structure prediction model based on machine learning and training it, and obtaining a building structure prediction model based on machine learning that meets a preset accuracy; wherein the building features include building geometry features, building shape features, and building points of interest (POI) features.

[0215] Wherein step S3 comprises the following steps:

[0216] Step S301: extract building features; and calculate building feature values;

[0217] Step S302: Evaluate and screen building characteristics;

[0218] Step S303: Determine a machine learning algorithm and establish a building structure prediction model based on machine learning;

[0219] Step S304: training a building structure prediction model based on machine learning;

[0220] Step S305: Evaluate the prediction results of the machine learning-based building structure prediction model to obtain a machine learning-based building structure prediction model that meets a preset accuracy.

[0221] Wherein, step S302 of evaluating and screening building characteristics includes:

[0222] Use the feature scoring function to score each feature;

[0223] Sort the features by their ratings from high to low.

[0224] Select features that rank in the top K of the preset positions.

[0225] In step S301, the calculation of the building characteristic value specifically includes:

[0226] The calculation of the building geometric characteristic value includes the following steps:

[0227] Predefine the number of building floors and building height;

[0228] The height between floors is calculated using formula (1):

[0229]

[0230] Use the geometry calculation function in ArcGIS platform to calculate the outline area and perimeter of the building;

[0231] The building outline is simplified into the minimum bounding rectangle MBR surrounding the building, and the length and width of the minimum bounding rectangle MBR are calculated by the minimum bounding rectangle geometry function of ArcGIS; the parameter is set to geometry type = Rectangle_By_WIDTH.

[0232] The length of the minimum bounding rectangle MBR is used to represent the true length of the building;

[0233] The width of the minimum bounding rectangle MBR is weighted and then used using formula (2) to represent the true width of the building outline:

[0234]

[0235] Calculate the ratio of the building's height to width and its length to width:

[0236] The building height to width ratio Rh2w is calculated using formula (3):

[0237]

[0238] The true length-to-width ratio of a building, Rl2w, is calculated using formula (4):

[0239]

[0240] Among them, H f is the height between floors, H b is the building height, N f is the number of floors of the building; W b is the true width of the building outline, A b is the outline area of ​​the building, A MBR is the area of ​​the minimum bounding rectangle MBR surrounding the building outline, L b is the actual length of the building, L MBR and W MBR are the length and width of the minimum bounding rectangle MBR respectively.

[0241] The building shape characteristic value of the building is calculated using formula (5):

[0242]

[0243] Among them, SI is the shape irregularity of the building outline. Since the MBR area is always greater than or equal to the building area, the value range is (0, 1.0], and the lower the value, the higher the irregularity.

[0244] The calculation is done using the ArcGIS "Feature Vertices To Points" function with the parameter Point_Type set to ALL. This function splits the building outline into a series of line segments and points, and the number of vertices is the total number of these points.

[0245] The calculation of the building point of interest POI feature value includes the following steps:

[0246] Count the total number of building points of interest POIs and the total number of building points of interest POI types for each building;

[0247] The building point of interest POI type is evaluated step by step using a decision tree until a leaf node is reached. The leaf node indicates the building point of interest POI type, and the corresponding Boolean value is obtained according to the building point of interest POI type.

[0248] The POI patterns of buildings in the training data are identified through a decision tree (maximum depth 9). The POIs of buildings are evaluated step by step along the decision tree until a leaf node is reached, which indicates the main POI type.

[0249] The string value of the major POI type is then converted to a numerical feature using binary one-hot encoding, creating a Boolean feature for each possible value of the major POI type. For example, if majorPOIType has two values ​​010000 and 020000, two Boolean features are generated: majorPOIType=010000 and majorPOIType=020000, each with a value of 0 or 1.

[0250] Finally, the majorPOIType feature has 16 major POI types, which are converted into 16 Boolean features.

[0251] In step S305, the model is evaluated using multiple indicators, including precision, recall, and F1 value;

[0252] Precision:

[0253]

[0254] Where: P i is the precision of category i; tpi is the number of buildings correctly identified as category i (true positives); fpi is the number of buildings incorrectly identified as category i (false positives);

[0255] Recall:

[0256]

[0257] Where: R i is the recall rate of category i; fn i is the number of buildings that should be identified as class i but were not (false negatives);

[0258] F1 value:

[0259]

[0260] Among them: F1 i is the F1 value of category i, which measures the harmonic mean of precision and recall.

[0261] The present invention is a single-label multi-classification problem (ie, each building can only belong to one category), so macro-based precision, recall and F1 value are used to ensure that each category has equal weight.

[0262] The formula for macro-average is as follows:

[0263] 1. Macro Precision:

[0264]

[0265] C is the total number of categories.

[0266] 2. Macro Recall:

[0267]

[0268] 3. Macro average F1 value (Macro F1):

[0269]

[0270] Step S4: Establishing a deep learning-based building material prediction model: Performing data enhancement on the material classification model data set, based on the YOLOV11 model, using the SwinTransformerV2 or ConvNeXtV2 computer vision network to replace the feature extraction part of the YOLOV11 model, establishing a deep learning-based building material prediction model and training it, and obtaining a deep learning-based building material prediction model that meets the preset accuracy;

[0271] Wherein, step S4 comprises the following steps:

[0272] Step S401: annotate the street view image (such as material standards, door and window annotations, and viewpoint classification) to obtain annotated material classification model dataset (dataset 1), and perform data enhancement using random cropping, rotation, tilting, and noise blurring techniques;

[0273] Random Cropping: Randomly crop some areas from the original image and enlarge them to the original size, so that the model can focus on different areas.

[0274] Rotation: Rotate the image by a random angle so that the model can adapt to changes in the image's orientation.

[0275] Tilt: Use an affine transformation to adjust the tilt of the image to account for different shooting angles.

[0276] Noise and blur: Add Gaussian noise or perform Gaussian blur to improve the robustness of the model.

[0277] Step S402: Based on the YOLOV11 model, the feature extraction part of the YOLOV11 model is replaced by the SwinTransformerV2 or ConvNeXtV2 computer vision network to establish a building material prediction model based on deep learning;

[0278] YOLO V11 is selected as the base model. YOLO V11 is an advanced target detection model known for its efficient detection capabilities and low computational requirements. YOLO divides the image into grids and predicts the object category and location in each grid, thereby achieving fast and efficient detection.

[0279] The target detection formula is as follows:

[0280] Loss total =Loss coord +Loss confidence +Loss class

[0281] in:

[0282] Loss coord is the location loss, which evaluates the difference between the predicted bounding box and the true bounding box.

[0283] Loss confidence It is the confidence loss, indicating whether the predicted box contains an object.

[0284] Loss class It is the category loss, the loss of the object category in the predicted box.

[0285] In order to improve the feature extraction capability of the YOLO V11 model, its feature extraction module is replaced with two visual networks, SwinTransformer V2 or ConvNeXt V2:

[0286] Swin Transformer V2: It uses a multi-head self-attention mechanism to divide the image into non-overlapping windows and perform self-attention calculations in each window, allowing the model to capture a wider range of features. The formula is as follows:

[0287]

[0288] Among them, Q (query), K (key) and V (value) are matrices obtained by transforming the image block, d k is the scaling factor.

[0289] ConvNeXt V2: It is an improved structure based on the convolutional network, using deeper convolutional layers and improved convolution kernel size to enhance the learning ability of features. The convolution calculation formula is:

[0290]

[0291] Among them, O(i, j) is the pixel value of the output image, I(i+m, j+n) is the pixel value of the input image, and K(m, n) is the convolution kernel.

[0292] Step S403: Optimizing the hyperparameters of the deep learning-based building material prediction model to select the best hyperparameters;

[0293] Step S404: training the deep learning-based building material prediction model after the hyperparameter optimization;

[0294] Step S405: Evaluate the trained deep learning-based building material prediction model to obtain a deep learning-based building material prediction model that meets a preset accuracy.

[0295] In step S405, the model is evaluated using precision, recall, average precision and F1 score as evaluation indicators;

[0296] The evaluation of the model uses precision, recall, mean average precision (mAP) and F1 score as indicators. The formula is as follows:

[0297] Precision:

[0298]

[0299] Among them, TP (True Positives) is the number of correctly predicted positive samples, and FP (False Positives) is the number of incorrectly predicted positive samples.

[0300] Recall:

[0301]

[0302] Among them, FN (False Negatives) is the number of samples that are actually positive but predicted to be negative.

[0303] Mean Average Precision (mAP): mAP is the average precision of all categories. For the precision and recall curves of each category c, calculate the average precision (AP) at different thresholds and then average it for all categories:

[0304]

[0305] F1 score: The F1 value is the harmonic average of precision and recall, and the formula is as follows:

[0306]

[0307] These indicators can comprehensively measure the classification performance of the model. By comparing the precision, recall, mAP and F1 value, the recognition accuracy and generalization ability of the model for building materials are evaluated.

[0308] 2. Use the building structure prediction model and building material prediction model to calculate carbon emissions based on the current characteristics of the building:

[0309] Step S5: Collect the status data of the building to be evaluated, obtain the building structure characteristic value and the building material characteristic value respectively through the building structure prediction model based on machine learning and the building material prediction model based on deep learning, and obtain the building status characteristic value;

[0310] Wherein, step S5 comprises the following steps:

[0311] Step S501: collecting current feature values ​​of building features through a geographic information system, including building geometric feature values, building shape feature values, and building point of interest POI feature values;

[0312] Step S502: Collecting street view images of the current status of buildings through online maps;

[0313] Step S503: The machine learning-based building structure prediction model and the deep learning-based building material prediction model obtain the building structure characteristic value and the building material characteristic value respectively, and obtain the building status characteristic value; the building status characteristic value includes the building structure category, the building material ratio, the building geometry characteristic value, the building shape characteristic value, and the building point of interest POI characteristic value;

[0314] Step S504: sorting out building structure categories, building material proportions, building geometric feature values, building shape feature values, and building point of interest POI feature values.

[0315] Among them, the building structure characteristic values ​​obtained through the building structure prediction model based on machine learning include the building structure types identified by the building structure prediction model based on machine learning, and the building structure volume Vmain body is calculated through street view image data. The total volume of the building is determined according to the proportion of the building structure volume in different structure types, different number of building floors and heights.

[0316] The main reference material “Building Structure” was used to obtain the proportion of building structures in different structures, different numbers of floors and heights.

[0317] Frame structure: The structural volume accounts for about 10%-15% of the total volume;

[0318] Shear wall structure: the structural volume accounts for 15%-20% of the total volume;

[0319] Steel structure: The structural volume accounts for 8%-12% of the total volume;

[0320] Low-rise buildings (1-3 floors): The structural volume accounts for a small proportion, usually around 10%-15%;

[0321] Medium- and high-rise buildings (4-20 floors): the structural volume accounts for between 15% and 20%;

[0322] Super high-rise buildings (more than 20 floors): the structural volume may account for 20%-25%;

[0323] Step S6: constructing a carbon emission calculation model based on building structure characteristics, building material characteristics and corresponding carbon emission factors;

[0324] Wherein step S6 comprises the following steps:

[0325] Step S601: Collect existing data to obtain carbon emission factors C of different structures 主体 and carbon emission factors C of different materials 维护 ;

[0326] Step S602: Obtaining the area and volume of the building through the building geometric feature values;

[0327] Step S603: Construct a building structure carbon emission model and a building material carbon emission model. According to the area and volume of the building, combined with the carbon emission coefficient corresponding to its structure type and material, calculate the carbon emission of the main structure and the surrounding structure of the building. The carbon emission calculation formula is as follows:

[0328]

[0329] in:

[0330] V 主体i : The volume of the i-th structure (m3 );

[0331] C 主体i : Carbon emission standard value of the i-th structure (kgCO 2 / m 3 );

[0332] α i : Carbon emission coefficient of the i-th structure, used to adjust the impact of different structures;

[0333] A 围护j : Surface area of ​​the jth material (m 2 );

[0334] C 围护j : Carbon emission standard value of the jth material (kgCO 2 / m 2 );

[0335] β j : Carbon emission coefficient of the jth material, used to consider the characteristics of different materials;

[0336] E 其他 :Carbon emissions from other construction processes (kgCO 2 );

[0337] E 总 :Total carbon emissions (kgCO 2 );

[0338] Step S604: Evaluate the building structure carbon emission model and the building material carbon emission model;

[0339] Evaluate whether the calculation results are reasonable from the following four aspects

[0340] Compare industry standards and reference values:

[0341] Compare the calculated carbon emissions with the industry average or reference values ​​from existing studies to determine the rationality of the calculated results. For example, you can compare it with the carbon emission standards of similar buildings or data from related studies.

[0342] Outlier Detection:

[0343] Check whether there are outliers in the calculation results (such as abnormally large carbon emissions), and analyze whether there are deviations in data input and model calculations. Statistical analysis can be performed on the distribution of carbon emissions to identify and explain possible unreasonable data.

[0344] Sensitivity analysis:

[0345] Conduct sensitivity analysis on key parameters (such as carbon emission coefficient, building life, etc.) to determine the sensitivity of the results to parameter changes, thereby evaluating the robustness of the model and the rationality of the results.

[0346] Expert Assessment:

[0347] Invite experts in construction, carbon emissions or environmental science to review and evaluate the model results to ensure that the results are consistent with the reality of building carbon emissions.

[0348] Step S605: Obtain the final carbon emission calculation model.

[0349] Step S7: Obtain an estimated value of the building's carbon emissions based on the building's current characteristic values ​​and the carbon emissions calculation model.

[0350] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and inference, characterized in that: The steps include: Step S1: Collect data, collect structure classification model data, including building data and building point of interest POI data, and collect material classification model data, including street view images; Step S2: pre-training data processing: for the structure classification model data, the buildings are labeled and matched with the points of interest to obtain the structure classification model data set; for the material classification model data, the data is segmented and labeled to obtain the material classification model data set; Step S3: Establishing a machine learning-based building structure prediction model: extracting building features and calculating building feature values, evaluating and screening building features, determining a machine learning algorithm, establishing a machine learning-based building structure prediction model and training it, and obtaining a machine learning-based building structure prediction model that meets a preset accuracy; Step S4: Establishing a deep learning-based building material prediction model: Performing data enhancement on the material classification model data set, based on the YOLOV11 model, using the SwinTransformerV2 or ConvNeXtV2 computer vision network to replace the feature extraction part of the YOLOV11 model, establishing a deep learning-based building material prediction model and training it, and obtaining a deep learning-based building material prediction model that meets the preset accuracy; Step S5: Collect the status data of the building to be evaluated, obtain the building structure characteristic value and the building material characteristic value respectively through the building structure prediction model based on machine learning and the building material prediction model based on deep learning, and obtain the building status characteristic value; Step S6: constructing a carbon emission calculation model based on building structure characteristics, building material characteristics and corresponding carbon emission factors; Step S7: Obtain an estimated value of the building's carbon emissions based on the building's current characteristic values ​​and the carbon emissions calculation model.

2. The method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and inference according to claim 1 is characterized in that: Step S1 includes the following steps: Step S101: Obtain building data including building footprints, outlines, and attributes through a geographic information system, and collect building point of interest (POI) data through a network map; Step S102: Collect street view images through real photos and online maps.

3. The method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and inference according to claim 1 is characterized in that: Step S2 includes the following steps: Step S201: Processing the structural classification model data, manually marking buildings with reference to building codes and technical standards, building-related publications, online satellite and street view maps, matching building data with point of interest (POI) data, and collating to obtain a complete structural classification data set in shp format; Step S202: Use roboflow to segment and annotate the street view image to obtain an annotated material classification model dataset, and perform data enhancement on the annotated material classification model dataset; Step S203: Divide the structure classification dataset and the street view dataset into a training set and a test set.

4. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to claim 3 is characterized in that: The labeled material classification model dataset is enhanced using techniques including random cropping, rotation, tilting, and noise blurring.

5. The method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and inference according to claim 1 is characterized in that: Step S3 includes the following steps: Step S301: extracting building features; And calculate the building characteristic value; Step S302: Evaluate and screen building characteristics; Step S303: Determine a machine learning algorithm and establish a building structure prediction model based on machine learning; Step S304: training a building structure prediction model based on machine learning; Step S305: Evaluate the prediction results of the machine learning-based building structure prediction model to obtain a machine learning-based building structure prediction model that meets a preset accuracy.

6. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to claim 5 is characterized in that: Assessment and screening of building characteristics include: Use the feature scoring function to score each feature; Sort the characteristics from high to low based on their ratings; Select features that rank in the top K of the preset positions.

7. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to any one of claims 1 to 6, characterized in that: Building features include building geometry features, building shape features, and building point of interest (POI) features.

8. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to claim 7 is characterized in that: The calculation of building geometric characteristic values ​​includes the following steps: Predefine the number of building floors and building height; The height between floors is calculated using formula (1): Use the geometry calculation function in ArcGIS platform to calculate the outline area and perimeter of the building; Simplify the building outline into the minimum bounding rectangle MBR surrounding the building, and calculate the length and width of the minimum bounding rectangle MBR through the minimum bounding geometry function of ArcGIS; The length of the minimum bounding rectangle MBR is used to represent the true length of the building; The width of the minimum bounding rectangle MBR is weighted and then used using formula (2) to represent the true width of the building outline: Calculate the ratio of the building's height to width and its length to width: The building height to width ratio Rh2w is calculated using formula (3): The true length-to-width ratio of a building, Rl2w, is calculated using formula (4): Among them, H f is the height between floors, H b is the building height, N f is the number of floors of the building; W b is the true width of the building outline, A b is the outline area of ​​the building, A MBR is the area of ​​the minimum bounding rectangle MBR surrounding the building outline, L b is the actual length of the building, L MBR and W MBR are the length and width of the minimum bounding rectangle MBR respectively.

9. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to claim 8 is characterized in that: The building shape characteristic value of the building is calculated using formula (5): Among them, SI is the shape irregularity of the building outline, and its value range is (0, 1.0]. The lower the value, the higher the irregularity.

10. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to claim 7 is characterized in that: The calculation of building point of interest POI feature value includes the following steps: Count the total number of building points of interest POIs and the total number of building points of interest POI types for each building; The building point of interest POI type is evaluated step by step using a decision tree until a leaf node is reached. The leaf node indicates the building point of interest POI type, and the corresponding Boolean value is obtained according to the building point of interest POI type.

11. The method for estimating the carbon footprint of a building representation based on artificial intelligence recognition and inference according to claim 1 is characterized in that: The machine learning algorithm identified is gradient boosted decision tree.

12. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to claim 1 is characterized in that: The building structure characteristic values ​​obtained through the building structure prediction model based on machine learning include the building structure types identified by the building structure prediction model based on machine learning, and the building structure volume Vmain body is calculated through street view image data. The total volume of the building is determined according to the proportion of the building structure volume in different structure types, different number of building floors and heights.

13. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to claim 1 is characterized in that: Step S4 includes the following steps: Step S401: annotate the street view image to obtain annotated material classification model data set, and use random cropping, rotation, tilting, and noise blurring techniques to enhance the data; Step S402: Based on the YOLOV11 model, the feature extraction part of the YOLOV11 model is replaced by the SwinTransformerV2 or ConvNeXtV2 computer vision network to establish a building material prediction model based on deep learning; Step S403: Optimizing the hyperparameters of the deep learning-based building material prediction model to select the best hyperparameters; Step S404: training the deep learning-based building material prediction model after the hyperparameter optimization; Step S405: Evaluate the trained deep learning-based building material prediction model to obtain a deep learning-based building material prediction model that meets a preset accuracy.

14. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to claim 7 is characterized in that: Step S5 includes the following steps: Step S501: collecting current feature values ​​of building features through a geographic information system, including building geometric feature values, building shape feature values, and building point of interest POI feature values; Step S502: Collecting street view images of the current status of buildings through online maps; Step S503: The machine learning-based building structure prediction model and the deep learning-based building material prediction model respectively obtain the building structure characteristic value and the building material characteristic value, and obtain the building status characteristic value; Step S504: sorting out building structure categories, building material proportions, building geometric feature values, building shape feature values, and building point of interest POI feature values.

15. The method for estimating the carbon footprint of a building representation based on artificial intelligence identification and inference according to claim 14 is characterized in that: Step S6 includes the following steps: Step S601: Collect existing data to obtain carbon emission factors C of different structures 主体 and carbon emission factors C of different materials 维护 ; Step S602: Obtaining the area and volume of the building through the building geometric feature values; Step S603: Construct a building structure carbon emission model and a building material carbon emission model. According to the area and volume of the building, combined with the carbon emission coefficient corresponding to its structure type and material, calculate the carbon emission of the main structure and the surrounding structure of the building. The carbon emission calculation formula is as follows: in: V 主体i : The volume of the i-th structure (m 3 ); C 主体i : Carbon emission standard value of the i-th structure (kgCO2 / m 3 ); α i : Carbon emission coefficient of the i-th structure, used to adjust the impact of different structures; A 围护j : Surface area of ​​the jth material (m 2 ); C 围护j : Carbon emission standard value of the jth material (kgCO2 / m 2 ); β j : Carbon emission coefficient of the jth material, used to consider the characteristics of different materials; E 其他 : Carbon emissions generated in other construction processes (kgCO2); E 总 : Total carbon emissions (kgCO2); Step S604: Evaluate the building structure carbon emission model and the building material carbon emission model; Step S605: Obtain the final carbon emission calculation model.

16. A building characterization carbon footprint estimation device based on artificial intelligence recognition and inference, characterized in that: It includes a processor, which executes the steps of the method for estimating the carbon footprint of building characterization based on artificial intelligence identification and speculation as described in any one of claims 1 to 15.

Images