Machine learning based linear cultural heritage visual prominence measure method and system

Abstract

The application discloses a kind of linear cultural heritage visual prominence measure method and system based on machine learning, the method comprises the following steps: first, construct geographic information space model and carry out spatial quantification analysis, obtain the visual prominence measure result based on computer simulation.Second, through eye tracker experiment and questionnaire survey, obtain the visual prominence measure result based on subjective perception.Then analyze the visual prominence measure result deviation obtained by two methods and establish the corresponding relationship between deviation data and block space characteristics.Then extract quantifiable block space scale characteristic index from four dimensions.Finally, construct the prediction deviation value model based on machine learning and correct and optimize the measurement result.The application overcomes the problems existing in the prior art, such as low measurement accuracy, large error, low repeatability and limited application scenarios, and provides solid data support and scientific basis for heritage protection decision-making.

Description

Technical Field

[0001] This invention relates to the field of heritage conservation technology, and in particular to a method and system for measuring the visual prominence of linear cultural heritage based on machine learning. Background Technology

[0002] Linear cultural heritage refers to a group of heritage sites composed of both tangible and intangible remains within a linear or strip-shaped area. Common forms include canals, roads, and railway lines, and it is characterized by its long historical span, wide geographical scope, and diverse heritage types. As a crucial means for people to directly perceive linear cultural heritage, the cultural charm and historical significance of this type of heritage in urban spaces are primarily revealed through visual perception.

[0003] Currently, methods for measuring the visual prominence of linear cultural heritage can be mainly categorized into two types: computer vision simulation and perceptual experience experiments. However, both methods, when applied individually to measure the visual prominence of linear cultural heritage, struggle to fully address its inherent complexity. Computer vision simulation, as an efficient assessment tool, can rapidly process large-scale spatial data, but its simulation results may deviate from actual human visual perception. Perceptual experience experiments, encompassing various subjective perception data collection methods such as eye-tracking, EEG experiments, and interviews, can directly measure the visual prominence of linear cultural heritage, effectively compensating for the shortcomings of pure geometric visual analysis. However, perceptual experience experiments typically rely on participant involvement, and are constrained by high time costs, limited experimental intensity, and difficulties in sample organization during experimental design and implementation. This makes it difficult to achieve high-density, multi-viewpoint comprehensive coverage within a large-scale continuous spatial range, thus presenting significant limitations when applied to measuring the visual prominence of linear cultural heritage.

[0004] Furthermore, due to the characteristics of linear cultural heritage, such as long distances across regions, complex heritage types, significant differences in spatial environments, and a large number of viewpoints, the factors affecting visual prominence and their mechanisms of action vary significantly across different spatial segments. This difference stems not only from the difficulty of computer simulations in fully replicating real human visual perception, but also from the segmental variations in spatial characteristics such as development intensity, river channel spatial relationships, distribution of heritage elements, and vegetation environment, which can easily lead to systematic biases in the results of the two assessment methods.

[0005] Therefore, the invention of a linear method and system for measuring the visual prominence of cultural heritage based on machine learning is an urgent need in the field of heritage conservation practice. Summary of the Invention

[0006] Technical Problem: To address the shortcomings of existing technologies, this invention provides a machine learning-based method and system for measuring the visual prominence of linear cultural heritage. The method constructs a multi-source visual prominence measurement approach that combines computer simulation and subjective perception, quantifies the deviation between the two in terms of visual prominence, and establishes a deviation prediction model based on spatial features. This model is used to correct the computer simulation results, making them highly approximate the true level of subjective perception, thereby achieving high-precision measurement of the visual prominence of linear cultural heritage while ensuring assessment efficiency. Furthermore, we propose a measurement system that, with the help of relevant computer technologies, makes the measurement of the visual prominence of linear cultural heritage more convenient, reliable, and effective. This invention provides a solid scientific basis and reliable technical support for heritage protection assessment and urban spatial planning.

[0007] Technical Solution: This invention provides a method for measuring the visual prominence of linear cultural heritage based on machine learning, specifically including the following steps:

[0008] Step 1: Construct a geographic information spatial model and perform spatial quantitative analysis to obtain visual salience measurement results based on computer simulation;

[0009] Step 2: Obtain visual assertiveness measurement results based on subjective perception through eye-tracking experiments and surveys;

[0010] Step 3: Analyze the deviation between the two visual responsibility measurement results and establish the correspondence between the visual responsibility deviation data and the spatial characteristics of the street blocks;

[0011] Step 4: Extract quantifiable neighborhood elements from four dimensions: development intensity, waterway elements, heritage elements, and vegetation elements.

[0012] Spatial scale characteristic indicators;

[0013] Step 5: Build a machine learning-based prediction bias model and correct and optimize the measurement results.

[0014] in,

[0015] Step 1 includes the following steps:

[0016] Step 1.1: Collect basic geographic information data of the linear cultural heritage area based on open source data information, import it into geographic information system software, construct a digital terrain spatial model of the city containing geographic information elements such as urban buildings, water bodies, topography and roads, and establish a unified coordinate system;

[0017] Step 1.2: Select and classify linear cultural heritage resource points according to the preset heritage value evaluation criteria, and construct cultural heritage resource points of different levels into independent data layers and load them into the spatial model generated in Step 1.1;

[0018] Step 1.3: Analyze the visibility of linear cultural heritage resource points using geographic information system software and programming tools. Quantify the analysis from two dimensions: the number of visible heritage resource points and the visible area, and obtain the measurement results of the visual prominence of linear cultural heritage based on computer simulation.

[0019] Step 2 includes the following steps:

[0020] Step 2.1: Use drones and handheld action cameras to collect on-site images along the linear cultural heritage route, obtain image data under different spatial locations and perspectives, and construct an image dataset;

[0021] Step 2.2: Based on this, an eye-tracking experiment is conducted. An eye tracker is used to record the gaze trajectory and gaze behavior data of the subjects during the process of viewing the image. The raw eye-tracking data is processed by denoising, time truncation and outlier removal. Visual behavior indicators such as the number of gazes, total gaze duration, average gaze duration and first gaze time are extracted to form a visual behavior indicator dataset.

[0022] Step 2.3: Conduct an offline survey to obtain subjective evaluation data on the visual prominence of linear cultural heritage; perform anomaly removal on the questionnaire data, and extract indicators such as heritage perception rate, perception intensity, and element perception ratio to form a subjective evaluation indicator dataset.

[0023] Step 2.4: Standardize the visual behavior index data and subjective evaluation index data, and use the AHP-CRITIC combined weighting method to determine the weight of each index, so as to obtain the linear cultural heritage visual prominence measurement results based on subjective perception.

[0024] Step 3 includes the following steps:

[0025] Step 3.1: Import the results of the linear cultural heritage visual prominence measurement based on computer simulation and the results of the linear cultural heritage visual prominence measurement based on subjective perception into the geographic information system, and establish the correspondence between the two types of measurement results through spatial matching.

[0026] Step 3.2: Perform Min-Max normalization on the two types of matched data, and then... The visual salience deviation value D between the computer simulation results and the subjective perception results at the same observation point is calculated to characterize the degree of inconsistency between the two types of measurement results. This represents the normalized computer-simulated visual salience measurement result. This represents the normalized result of the subjective perceived visual assertiveness measurement.

[0027] Step 3.3: Rationally divide the spatial analysis units into street block scales, import the street block geographic information vector data, and use spatial overlay analysis tools to establish the spatial correlation between the visual prominence deviation data values ​​and the corresponding street block units, forming a deviation attribute database based on street blocks.

[0028] Step 4 includes the following steps:

[0029] Step 4.1: Construct a multi-dimensional and quantifiable spatial characteristic index system at the block scale from four aspects: development intensity, river elements, heritage elements, and vegetation elements. Among them, development intensity represents the level of spatial development of the built environment of different blocks, river elements represent the spatial relationship between river water and the built environment of different blocks, heritage elements represent the degree of spatial aggregation or dispersion of linear cultural heritage resource points, and vegetation elements represent the characteristics of spatial vegetation environment.

[0030] Step 4.2: Based on the urban digital terrain spatial model generated in Step 1.1, spatial geometric calculation and statistical analysis are performed using a geographic information system to obtain development intensity characteristic indicators, including the ratio of street perimeter to area, building density, and plot ratio.

[0031] Step 4.3: Extract water body boundary data from the urban digital terrain spatial model and obtain river feature index results through spatial analysis tools; among them, calculate the shortest and farthest distances between blocks and water bodies through spatial distance analysis tools, and calculate the elevation difference between blocks and rivers through terrain elevation overlay analysis.

[0032] Step 4.4: Extract vector data of linear cultural heritage resource points from the urban digital terrain spatial model. Using the kernel density analysis tool in the geographic information system, calculate the kernel density of heritage resource points of different levels to generate a rasterized density map of continuous spatial distribution, which is used to characterize the spatial clustering characteristics of heritage resource points. Call the average nearest neighbor analysis tool to obtain the nearest neighbor index of heritage resource points, which is used to characterize the degree of clustering or dispersion of heritage resource points in space, thus completing the extraction and expression of spatial characteristic indicators of heritage elements.

[0033] Step 4.5: Obtain and import basic vegetation data. Based on the vegetation spatial data, statistically obtain indicators such as green space ratio, average vegetation height, and vegetation green volume index to complete the extraction and expression of vegetation element characteristic indicators.

[0034] The formula for calculating the ratio of street perimeter to area is:

[0035] ,

[0036] in This represents the ratio of the perimeter of the buildings in the i-th block. This indicates the perimeter of the block. Indicates the area of ​​the block;

[0037] Block building density represents the proportion of building footprint area in the total area of ​​a block, and its calculation formula is as follows:

[0038] ,

[0039] in This represents the building density of the i-th block. This indicates the number of buildings in the block. Indicates the i-th block. The floor area of ​​each building, Indicates the area of ​​the block

[0040] The floor area ratio (FAR) of a neighborhood reflects the vertical development intensity of a neighborhood, and its calculation formula is as follows:

[0041] ,

[0042] in This represents the floor area ratio of the i-th block. This indicates the number of buildings in the block. Indicates the i-th block. The total area of ​​the buildings This indicates the area of ​​the block.

[0043] The formula for calculating the green space ratio is:

[0044] ,

[0045] in This represents the green space ratio of the i-th block. This indicates the area of ​​the block. This represents the green space area of ​​the i-th block;

[0046] The formula for calculating the area of ​​vegetation green coverage index is:

[0047] ,

[0048] Where r and h represent the average canopy projection radius and average height of vegetation in the i-th block, respectively, and C i This represents the green space ratio of the i-th block.

[0049] Step 5 includes the following steps:

[0050] Step 5.1: Using the visual prominence deviation data obtained in Step 3 as the dependent variable and the street-scale spatial feature indicators obtained in Step 4 as the independent variables, construct a machine learning modeling dataset and divide it into a training set and a test set according to a preset ratio of 8:2. The training set is used for model training and the test set is used for model performance verification to obtain standardized modeling data input.

[0051] Step 5.2: In the Python environment, call scikit-learn and related machine learning libraries to build various visual salience bias prediction models, including decision tree model, random forest model, gradient boosting tree model, XGBoost model and LightGBM model; input the training set data into each model for training, and obtain the bias value prediction function through model fitting;

[0052] Step 5.3: Simultaneously, the cross-validation function is invoked to perform 5-fold cross-validation to evaluate the stability and generalization ability of each model on different data subsets. Based on the prediction results on the test set, combined with the coefficient of determination R... 2 Evaluation metrics such as mean squared error (MSE) and mean absolute error (MAE) were used to compare and analyze the performance of different models, and the visual salience deviation prediction model with the best fitting effect was selected.

[0053] Step 5.4: Based on the optimal visual resilience deviation prediction model obtained in Step 5.3, input the spatial feature index data corresponding to each observation point for batch prediction to obtain the predicted visual resilience deviation value for each observation point; correct the original computer simulation visual resilience measurement results using the following formula:

[0054]

[0055] in The corrected computer simulation results for visual saliency. The results are the original computer simulation results. The visual prominence bias value predicted by the model;

[0056] Step 5.5 involves statistically comparing the corrected computer-simulated visual prominence results with the original computer-simulated results and subjective perception results. An error comparison analysis method is used to quantitatively evaluate the deviation between the corrected results and the subjective perception results, thus verifying the numerical optimization effect after deviation correction. The error formula is as follows:

[0057] , ,

[0058] Where Error_Before represents the actual error and Error_After represents the corrected error. This represents the normalized composite value from computer simulations. This represents the normalized composite value of subjective perception. The visual prominence bias value predicted by the model;

[0059] Step 5.5: Using spatial interpolation tools, the three types of data—the computer-simulated visual prominence results before and after correction, and the subjective perception results—are spatially continuous to generate corresponding raster distribution maps. By comparing the consistency changes between the spatial distribution before and after correction and the subjective perception spatial distribution, the effectiveness of the deviation correction model is verified and the measurement results of the linear cultural heritage visual prominence in the region are obtained.

[0060] The measurement system of the linear cultural heritage visual prominence measurement method based on machine learning includes a data acquisition module, a feature index extraction module, and a model construction and prediction correction module, and the modules of the system are interconnected.

[0061] The data acquisition module is used to acquire and process geographic information data related to linear cultural heritage areas, obtain visual prominence measurement results based on computer simulation and subjective perception, and establish the correspondence between the deviation data of the two visual prominence measurement results and the spatial characteristics of the blocks; acquire basic spatial data of linear cultural heritage, heritage resource point data and field-collected image data, and construct a digital terrain spatial model of the city under a unified spatial benchmark; conduct computer simulation visual evaluation based on the spatial model, and obtain subjective perception visual prominence measurement results by combining eye-tracking experiments and questionnaire surveys; then perform spatial matching and normalization processing on the two types of measurement results, and output the deviation data results of visual prominence of linear cultural heritage.

[0062] The feature index extraction module is used to construct multi-dimensional and quantifiable street spatial scale feature indexes. Based on the linear cultural heritage visual prominence deviation data results generated by the data acquisition module, and combined with the spatial feature information of the linear cultural heritage area, feature indexes affecting visual prominence deviation are extracted from four dimensions: development intensity, heritage elements, river elements and vegetation elements, and street spatial scale feature indexes are constructed to characterize the spatial environment features.

[0063] The model building and prediction correction module is used to build a visual responsibility deviation prediction model using machine learning methods and correct the prediction results. It takes visual responsibility deviation data and spatial scale feature index data generated by the input data acquisition module and the feature index extraction module as input, uses machine learning methods to build a visual responsibility deviation prediction model, predicts and evaluates the visual responsibility deviation of linear cultural heritage, and corrects the computer simulation visual responsibility measurement results based on the predicted deviation results, so that the corrected results are close to the actual perceived visual responsibility measurement results, thereby obtaining the optimized visual responsibility measurement results of linear cultural heritage and completing the visualization process.

[0064] Beneficial Effects: Compared with existing technologies, the beneficial effects of this invention are as follows: 1. This invention overcomes the problem that existing methods, due to the difficulty in simultaneously considering large-scale spatial morphology and real perception in a single assessment, lead to measurement bias between computer simulation and subjective perception. By constructing a multi-source visual assessment system and establishing a bias prediction model based on spatial features to correct simulation results, making them closer to real perception, this invention achieves high-precision measurement of the visual prominence of linear cultural heritage while ensuring efficiency, significantly improving the accuracy and reliability of the assessment. 2. This invention, by constructing a spatial feature index system at the street block scale and introducing machine learning methods to analyze the formation mechanism of measurement bias, can effectively reveal the complex nonlinear relationship between multi-dimensional spatial features such as development intensity, river spatial relationships, heritage element distribution, and vegetation environment and visual prominence bias. This clarifies the intrinsic mechanism of the difference between computer simulation and subjective perception, providing a reliable basis for bias prediction, result correction, and spatial optimization in the measurement of the visual prominence of linear cultural heritage. 3. This invention has good universality and scalability. It can flexibly adjust and optimize spatial characteristic indicators for different urban spatial forms and linear cultural heritage characteristics, thereby significantly improving the transferability and generalization ability of the whole set of evaluation methods in multiple scenarios, multiple scales and multiple types of linear heritage spaces. Attached Figure Description

[0065] Figure 1 This is a flowchart of the method and system described in this invention;

[0066] Figure 2 This is a schematic diagram of the research scope in the example;

[0067] Figure 3 The results of the visual salience measurement in the example are computer simulations.

[0068] Figure 4 The result is a subjectively perceived measure of visual prominence in the example.

[0069] Figure 5 This is a spatial mapping analysis diagram of the deviation values ​​between computer simulation and subjective perception in the example.

[0070] Figure 6 Framework diagram for constructing multi-model prediction models;

[0071] Figure 7 The example uses the learning predictions and linear fitting results of five machine models.

[0072] Figure 8 This is a set of comparative figures showing the spatial distribution of the measurement space in the example, based on computer simulation, subjective perception, and correction. Detailed Implementation

[0073] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art fall within the scope defined by the appended claims.

[0074] The technical method of the present invention will be described in detail below with reference to the accompanying drawings and embodiments. This embodiment takes the linear cultural heritage of the Grand Canal (Huai'an section) as the research object. This section is approximately 24 km long, and the research scope includes the main channel and the waterfront hinterland within a 1 km radius on both sides. This area has relatively typical pedestrian perception characteristics and visual continuity, and can be used as a research object based on visibility analysis, such as... Figure 2 As shown, the specific operation steps are as follows:

[0075] Step 1: Construct a geographic information spatial model and perform spatial quantitative analysis to obtain visual salience measurement results based on computer simulation, as follows:

[0076] Step 1.1: Open-source spatial data was obtained through public internet platforms, and BIGEMAP software was used for auxiliary extraction and preprocessing to obtain basic geographic information data such as building outlines, number of building floors, water system distribution, and road network. This data was then imported into the ArcGIS urban digital terrain spatial model, and the coordinate system of each data layer was unified to WGS 1984UTM Zone 50N. This study uses a spatial resolution of 10 m × 10 m to generate elevation raster data. This grid scale can effectively control the computation time and cost caused by large-scale view area calculations while ensuring the accuracy of urban spatial form representation.

[0077] Step 1.2, further, select and classify linear cultural heritage resource points according to the preset heritage value evaluation criteria, and establish independent layers for first-level, second-level and third-level heritage resource points in ArcGIS based on the classification results and load them into the spatial model generated in Step 1.1 to represent the spatial distribution characteristics of heritage elements of different value levels.

[0078] Step 1.3: Based on the constructed urban digital terrain spatial model, ArcGIS combined with Python scripts is used to quantitatively analyze the visibility of linear cultural heritage from two dimensions: the number of visible heritage resource points and the spatial area of ​​visible heritage. This yields computer-simulated visual prominence measurements for each viewpoint, and the results are output as shown in the figure. Figure 3 The spatial distribution results dataset is shown.

[0079] Step 2: Obtain visual assertiveness measurement results based on subjective perception through eye-tracking experiments and questionnaires, as follows:

[0080] Step 2.1: Using drones and handheld action cameras, on-site image collection was carried out along the linear cultural heritage within the research area to obtain image data under different spatial locations and perspectives, and an image dataset was constructed. A total of 16,330 valid images were obtained. The image collection points were evenly distributed at 10 m intervals to form a high-density continuous observation sequence.

[0081] Step 2.2: Based on this, an eye-tracking experiment was conducted, using an eye tracker to record the gaze trajectories and gaze behavior data of 35 subjects during image viewing. The acquired raw eye-tracking data was processed by denoising, time truncation, and outlier removal to extract visual behavior indicators such as the number of fixations, total fixation duration, average fixation duration, and first fixation time, forming a visual behavior indicator dataset.

[0082] Step 2.3: Further, an offline questionnaire survey was conducted within the linear cultural heritage area. This survey covered the entire spatial range of the linear cultural heritage, and a total of 805 valid questionnaires were collected to obtain subjective evaluation data on the visual prominence of the linear cultural heritage. Anomaly removal was performed on the questionnaire data, and indicators such as heritage perception rate, perception intensity, and element perception ratio were extracted to form a subjective evaluation index dataset.

[0083] Step 2.4 involves dimensionless processing of the index data obtained in Steps 2.1 and 2.2, and determining the weights of each index using the AHP-CRITIC combined weighting method. The visual salience evaluation results for each observation point based on subjective perception are obtained through a weighted summation method, such as... Figure 4 As shown.

[0084] Step 3: Analyze the discrepancy between the visual prominence assessment data obtained by the two methods and establish the correspondence between the discrepancy data and the spatial characteristics of the street blocks, as detailed below:

[0085] Step 3.1: Import the visual salience measurement results based on computer simulation and the visual salience measurement results based on subjective perception obtained in steps 1.3 and 2.4 into the ArcGIS platform, perform spatial matching processing under a unified coordinate system, and establish a one-to-one correspondence between the two types of measurement results.

[0086] Step 3.2: Perform Min-Max normalization on the two types of matched data, and then... The visual salience deviation value D between the computer simulation results and the subjective perception results at the same observation point is calculated to characterize the degree of inconsistency between the two types of measurement results, such as... Figure 5 As shown. This represents the normalized computer-simulated visual salience measurement result. This represents the normalized result of the subjective perceived visual assertiveness measurement.

[0087] Step 3.3: Rationally divide the spatial analysis units into street-level units and import the street geographic information vector data into ArcGIS. Use the Spatial Join tool to establish the spatial relationship between the visual visibility deviation data values ​​and the corresponding street units, forming a deviation attribute database based on street units.

[0088] Step 4: Extract quantifiable spatial scale characteristic indicators of the street block from four dimensions, as follows:

[0089] Step 4.1 constructs a multi-dimensional and quantifiable spatial characteristic index system at the block scale from four aspects: development intensity, river elements, heritage elements, and vegetation elements. Among them, development intensity represents the level of spatial development of the built environment of different blocks, river elements represent the spatial relationship between river water bodies and the built environment of different blocks, heritage elements represent the degree of spatial aggregation or dispersion of linear cultural heritage resource points, and vegetation elements represent the characteristics of spatial vegetation environment.

[0090] Step 4.2, based on the urban digital terrain spatial model generated in Step 1.1, uses a geographic information system to perform spatial geometric calculations and statistical analysis to obtain development intensity characteristic indicators, including the ratio of street perimeter to area, building density, and plot ratio.

[0091] (1) The formula for calculating the ratio of the perimeter to the area of ​​a street block is:

[0092] ,

[0093] in This represents the ratio of the perimeter of the buildings in the i-th block. This indicates the perimeter of the block. This indicates the area of ​​the block.

[0094] (2) The building density of a block represents the proportion of the building footprint area in the block's land area, and its calculation formula is as follows:

[0095] ,

[0096] in This represents the building density of the i-th block. This indicates the number of buildings in the block. Indicates the i-th block. The floor area of ​​each building, Indicates the area of ​​the block;

[0097] (3) The floor area ratio of a neighborhood is used to reflect the three-dimensional intensity of neighborhood development, and its calculation formula is as follows:

[0098] ,

[0099] in This represents the floor area ratio of the i-th block. This indicates the number of buildings in the block. Indicates the i-th block. The total area of ​​the buildings This indicates the area of ​​the block.

[0100] Step 4.3: Extract water body boundary data from the urban digital terrain spatial model and obtain river feature index results using spatial analysis tools. Specifically, the shortest and farthest distances between blocks and water bodies are calculated using spatial distance analysis tools. The elevation difference between blocks and the river is calculated using terrain elevation overlay analysis.

[0101] Step 4.4: Extract vector data of linear cultural heritage resource points from the urban digital terrain spatial model. Using the Kernel Density tool in ArcGIS, calculate the kernel density for heritage resource points of different levels to generate a rasterized density map of continuous spatial distribution, used to characterize the spatial clustering characteristics of heritage resource points. Further, use the Average Nearest Neighbor tool in the Spatial Statistics module of ArcGIS to obtain the nearest neighbor index of heritage resource points, used to characterize the degree of spatial clustering or dispersion of heritage resource points, completing the extraction and expression of spatial characteristic indicators of heritage elements.

[0102] Step 4.5: Obtain and import basic vegetation data. Based on the spatial vegetation data, statistically obtain indicators such as green space ratio, average vegetation height, and vegetation green volume index to complete the extraction and expression of vegetation element characteristic indicators.

[0103] (4) The formula for calculating the green space ratio is:

[0104] ,

[0105] in, This represents the green space ratio of the i-th block. This indicates the area of ​​the block. This represents the green space area of ​​the i-th block.

[0106] (5) The formula for calculating the area of ​​vegetation green volume index is:

[0107] ,

[0108] Where r and h represent the average canopy projection radius and average height of vegetation in the i-th block, respectively, and C i This represents the green space ratio of the i-th block.

[0109] Step 5: Construct a prediction bias model based on machine learning and perform corrections and optimizations, as follows:

[0110] Step 5.1: Using the visual salience deviation data obtained in Step 3 as the dependent variable, and the street-scale spatial characteristic indicators obtained in Step 4 as the independent variables, such as... Figure 6 As shown, a machine learning modeling dataset is constructed using a Python program and divided into a training set and a test set according to a preset ratio (8:2). The training set is used for model training, and the test set is used for model performance verification, resulting in standardized modeling data input.

[0111] Step 5.2: In the Python environment, call scikit-learn and related machine learning libraries to construct various visual salience bias prediction models, including decision tree, random forest, gradient boosting, XGBoost, and LightGBM models. Input the training set data into each model for training, and obtain the bias prediction function through model fitting.

[0112] Step 5.3 further involves simultaneously invoking the cross-validation function to perform 5-fold cross-validation, evaluating the stability and generalization ability of each model on different subsets of data. Based on the prediction results on the test set, combined with the coefficient of determination (R²),... 2 Evaluation metrics such as mean squared error (MSE) and mean absolute error (MAE) were used to compare and analyze the performance of different models, and the visual responsibility bias prediction model with the best fit was selected. In this embodiment, the visual responsibility bias prediction model with the best fit is the XGBoost model, such as... Figure 7 As shown

[0113] Step 5.4: Based on the optimal visual resilience deviation prediction model obtained in Step 5.3, input the spatial feature index data corresponding to each observation point for batch prediction to obtain the predicted visual resilience deviation value for each observation point. Further, the original computer-simulated visual resilience measurement results are corrected using the following formula:

[0114] ,

[0115] in The corrected computer simulation results for visual saliency. The results are the original computer simulation results. This represents the visual prominence deviation value predicted by the model.

[0116] Step 5.4 involves statistically comparing the corrected computer-simulated visual prominence results with the original computer-simulated results and subjective perception results in a Python program. Error comparison analysis is used to quantitatively evaluate the deviation between the corrected results and the subjective perception results, thus verifying the numerical optimization effect after deviation correction. The error formula is as follows:

[0117] ,

[0118] ,

[0119] Where Error_Before represents the true error, Error_After represents the corrected error, and V sim V represents the normalized composite value from computer simulations. pre This represents the normalized composite value of subjective perception. This represents the deviation value predicted by the model.

[0120] Step 5.5: Use the Kriging tool in ArcGIS's Spatial Analyst module to perform spatial continuity processing on the three types of data: the computer-simulated visual sharpness results before and after correction, and the subjective perception results, generating corresponding raster distribution maps. By comparing the consistency changes between the spatial distribution before and after correction and the subjective perception spatial distribution, such as... Figure 8 As shown, the effectiveness of the bias correction model is verified and the results of the linear cultural heritage visual prominence measurement in this region are obtained.

[0121] Furthermore, this invention proposes a linear cultural heritage visual prominence measurement system based on machine learning, comprising a data acquisition module, a feature index extraction module, and a model construction and prediction correction module, with each module interconnected.

[0122] (1) The data acquisition module is used to acquire and process geographic information data related to linear cultural heritage and the region, obtain visual prominence measurement results based on computer simulation and subjective perception, and establish the correspondence between the deviation data of the two visual prominence measurement results and the spatial characteristics of the street. Specifically, it acquires basic spatial data of linear cultural heritage, heritage resource point data and field-collected image data, and constructs a digital terrain spatial model of the city under a unified spatial benchmark. Based on the spatial model, it conducts computer simulation visual evaluation, and obtains subjective perception visual prominence measurement results by combining eye-tracking experiments and questionnaire surveys. Subsequently, it performs spatial matching and normalization processing on the two types of measurement results, and outputs the deviation data results of visual prominence of linear cultural heritage.

[0123] (2) The feature index extraction module is used to construct multi-dimensional quantifiable street spatial scale feature indicators. Specifically, based on the linear cultural heritage visual prominence deviation data results generated by the data acquisition module, and combined with the spatial feature information of the linear cultural heritage area, feature indicators affecting visual prominence deviation are extracted from four dimensions: development intensity, heritage elements, river elements and vegetation elements, and street spatial scale feature indicators are constructed to characterize the spatial environment features.

[0124] (3) The model construction and prediction correction module is used to construct a visual responsibility deviation prediction model using machine learning methods and to correct the prediction results. Specifically, the visual responsibility deviation data and spatial scale feature index data generated by the input data acquisition module and the feature index extraction module are used to construct a visual responsibility deviation prediction model using machine learning methods to predict and evaluate the visual responsibility deviation of linear cultural heritage. Based on the deviation results obtained from the prediction, the computer simulation visual responsibility measurement results are corrected so that the corrected results are close to the actual perceived visual responsibility measurement results, thereby obtaining the optimized visual responsibility measurement results of linear cultural heritage and completing the visualization process.

Claims

1. A method for measuring the visual prominence of linear cultural heritage based on machine learning, characterized in that, The method includes the following steps: Step 1: Construct a geographic information spatial model and perform spatial quantitative analysis to obtain visual salience measurement results based on computer simulation; Step 2: Obtain visual assertiveness measurement results based on subjective perception through eye-tracking experiments and surveys; Step 3: Analyze the deviation between the two visual responsibility measurement results and establish the correspondence between the visual responsibility deviation data and the spatial characteristics of the street blocks; Step 4: Extract quantifiable spatial scale characteristic indicators of the block from four dimensions: development intensity, river elements, heritage elements, and vegetation elements. Step 5: Build a machine learning-based prediction bias model and correct and optimize the measurement results.

2. The method for measuring the visual prominence of linear cultural heritage based on machine learning according to claim 1, characterized in that, Step 1 includes the following steps: Step 1.1: Collect basic geographic information data of the linear cultural heritage area based on open source data information, import it into geographic information system software, construct a digital terrain spatial model of the city containing geographic information elements such as urban buildings, water bodies, topography and roads, and establish a unified coordinate system; Step 1.2: Select and classify linear cultural heritage resource points according to the preset heritage value evaluation criteria, and construct cultural heritage resource points of different levels into independent data layers and load them into the spatial model generated in Step 1.1; Step 1.3: Analyze the visibility of linear cultural heritage resource points using geographic information system software and programming tools. Quantify the analysis from two dimensions: the number of visible heritage resource points and the visible area, and obtain the measurement results of the visual prominence of linear cultural heritage based on computer simulation.

3. The method for measuring the visual prominence of linear cultural heritage based on machine learning according to claim 2, characterized in that, Step 2 includes the following steps: Step 2.1: Use drones and handheld action cameras to collect on-site images along the linear cultural heritage route, obtain image data under different spatial locations and perspectives, and construct an image dataset; Step 2.2: Based on this, an eye-tracking experiment is conducted. An eye tracker is used to record the gaze trajectory and gaze behavior data of the subjects during the process of viewing the image. The raw eye-tracking data is processed by denoising, time truncation and outlier removal. Visual behavior indicators such as the number of gazes, total gaze duration, average gaze duration and first gaze time are extracted to form a visual behavior indicator dataset. Step 2.3: Conduct an offline survey to obtain subjective evaluation data on the visual prominence of linear cultural heritage; perform anomaly removal on the questionnaire data, and extract indicators such as heritage perception rate, perception intensity, and element perception ratio to form a subjective evaluation indicator dataset. Step 2.4: Standardize the visual behavior index data and subjective evaluation index data, and use the AHP-CRITIC combined weighting method to determine the weight of each index, so as to obtain the linear cultural heritage visual prominence measurement results based on subjective perception.

4. The method for measuring the visual prominence of linear cultural heritage based on machine learning according to claim 3, characterized in that, Step 3 includes the following steps: Step 3.1: Import the results of the linear cultural heritage visual prominence measurement based on computer simulation and the results of the linear cultural heritage visual prominence measurement based on subjective perception into the geographic information system, and establish the correspondence between the two types of measurement results through spatial matching. Step 3.2: Perform Min-Max normalization on the two types of matched data, and then... The visual salience deviation value D between the computer simulation results and the subjective perception results at the same observation point is calculated to characterize the degree of inconsistency between the two types of measurement results. This represents the normalized computer-simulated visual salience measurement result. This represents the normalized result of the subjective perceived visual assertiveness measurement. Step 3.3: Rationally divide the spatial analysis units into street block scales, import the street block geographic information vector data, and use spatial overlay analysis tools to establish the spatial correlation between the visual prominence deviation data values ​​and the corresponding street block units, forming a deviation attribute database based on street blocks.

5. The method for measuring the visual prominence of linear cultural heritage based on machine learning according to claim 4, characterized in that, Step 4 includes the following steps: Step 4.1: Construct a multi-dimensional and quantifiable spatial characteristic index system at the block scale from four aspects: development intensity, river elements, heritage elements, and vegetation elements. Among them, development intensity represents the level of spatial development of the built environment of different blocks, river elements represent the spatial relationship between river water and the built environment of different blocks, heritage elements represent the degree of spatial aggregation or dispersion of linear cultural heritage resource points, and vegetation elements represent the characteristics of spatial vegetation environment. Step 4.2: Based on the urban digital terrain spatial model generated in Step 1.1, spatial geometric calculation and statistical analysis are performed using a geographic information system to obtain development intensity characteristic indicators, including the ratio of street perimeter to area, building density, and plot ratio. Step 4.3: Extract water body boundary data from the urban digital terrain spatial model and obtain river feature index results through spatial analysis tools; among them, calculate the shortest and farthest distances between blocks and water bodies through spatial distance analysis tools, and calculate the elevation difference between blocks and rivers through terrain elevation overlay analysis. Step 4.4: Extract vector data of linear cultural heritage resource points from the urban digital terrain spatial model. Using the kernel density analysis tool in the geographic information system, calculate the kernel density of heritage resource points of different levels to generate a rasterized density map of continuous spatial distribution, which is used to characterize the spatial clustering characteristics of heritage resource points. Call the average nearest neighbor analysis tool to obtain the nearest neighbor index of heritage resource points, which is used to characterize the degree of clustering or dispersion of heritage resource points in space, thus completing the extraction and expression of spatial characteristic indicators of heritage elements. Step 4.5: Obtain and import basic vegetation data. Based on the vegetation spatial data, statistically obtain indicators such as green space ratio, average vegetation height, and vegetation green volume index to complete the extraction and expression of vegetation element characteristic indicators.

6. The method for measuring the visual prominence of linear cultural heritage based on machine learning according to claim 5, characterized in that, The formula for calculating the ratio of a street's perimeter to its area is: ， in This represents the ratio of the perimeter of the buildings in the i-th block. This indicates the perimeter of the block. Indicates the area of ​​the block; Block building density represents the proportion of building footprint area in the total area of ​​a block, and its calculation formula is as follows: ， in This represents the building density of the i-th block. This indicates the number of buildings in the block. Indicates the i-th block. The floor area of ​​each building, This indicates the area of ​​the block. The floor area ratio (FAR) of a neighborhood reflects the vertical development intensity of a neighborhood, and its calculation formula is as follows: ， in This represents the floor area ratio of the i-th block. This indicates the number of buildings in the block. Indicates the i-th block. The total area of ​​the buildings This indicates the area of ​​the block.

7. The method for measuring the visual prominence of linear cultural heritage based on machine learning according to claim 6, characterized in that, The formula for calculating the green space ratio is: ， in This represents the green space ratio of the i-th block. This indicates the area of ​​the block. This represents the green space area of ​​the i-th block; The formula for calculating the area of ​​vegetation green coverage index is: ， Where r and h represent the average canopy projection radius and average height of vegetation in the i-th block, respectively, and C i This represents the green space ratio of the i-th block.

8. The method for measuring the visual prominence of linear cultural heritage based on machine learning according to claim 7, characterized in that, Step 5 includes the following steps: Step 5.1: Using the visual prominence deviation data obtained in Step 3 as the dependent variable and the street-scale spatial feature indicators obtained in Step 4 as the independent variables, construct a machine learning modeling dataset and divide it into a training set and a test set according to a preset ratio of 8:

2. The training set is used for model training and the test set is used for model performance verification to obtain standardized modeling data input. Step 5.2: In the Python environment, call scikit-learn and related machine learning libraries to build various visual salience bias prediction models, including decision tree model, random forest model, gradient boosting tree model, XGBoost model and LightGBM model; input the training set data into each model for training, and obtain the bias value prediction function through model fitting; Step 5.3: Simultaneously, the cross-validation function is invoked to perform 5-fold cross-validation to evaluate the stability and generalization ability of each model on different subsets of data. Based on the prediction results on the test set, combined with the coefficient of determination R... 2 Evaluation metrics such as mean squared error (MSE) and mean absolute error (MAE) were used to compare and analyze the performance of different models, and the visual prominence deviation prediction model with the best fitting effect was selected. Step 5.4: Based on the optimal visual resilience deviation prediction model obtained in Step 5.3, input the spatial feature index data corresponding to each observation point for batch prediction to obtain the predicted visual resilience deviation value for each observation point; correct the original computer simulation visual resilience measurement results using the following formula: in The corrected computer simulation results for visual salience. The results are the original computer simulation results. The visual prominence bias value predicted by the model; Step 5.5 involves statistically comparing the corrected computer-simulated visual prominence results with the original computer-simulated results and subjective perception results. An error comparison analysis method is used to quantitatively evaluate the deviation between the corrected results and the subjective perception results, thus verifying the numerical optimization effect after deviation correction. The error formula is as follows: ， ， Where Error_Before represents the actual error and Error_After represents the corrected error. This represents the normalized composite value from computer simulations. This represents the normalized composite value of subjective perception. The visual prominence bias value predicted by the model; Step 5.5: Using spatial interpolation tools, the three types of data—the computer-simulated visual prominence results before and after correction, and the subjective perception results—are spatially continuous to generate corresponding raster distribution maps. By comparing the consistency changes between the spatial distribution before and after correction and the subjective perception spatial distribution, the effectiveness of the deviation correction model is verified and the measurement results of the linear cultural heritage visual prominence in the region are obtained.

9. A measurement system for the linear cultural heritage visual prominence measurement method based on machine learning as described in claim 8, characterized in that, The system includes a data acquisition module, a feature index extraction module, and a model building and prediction correction module, and the modules are interconnected. The data acquisition module is used to acquire and process geographic information data related to linear cultural heritage areas, obtain visual prominence measurement results based on computer simulation and subjective perception respectively, and establish the correspondence between the deviation data of the two visual prominence measurement results and the spatial characteristics of the blocks; acquire basic spatial data of linear cultural heritage, heritage resource point data and field-collected image data, and construct a digital terrain spatial model of the city under a unified spatial benchmark; Based on the spatial model, computer simulation visual assessment was conducted, and subjective perceived visual prominence measurement results were obtained by combining eye-tracking experiments and questionnaire surveys. Subsequently, spatial matching and normalization were performed on the two types of measurement results to output linear cultural heritage visual prominence deviation data results. The feature index extraction module is used to construct multi-dimensional and quantifiable street spatial scale feature indexes. Based on the linear cultural heritage visual prominence deviation data results generated by the data acquisition module, and combined with the spatial feature information of the linear cultural heritage area, feature indexes affecting visual prominence deviation are extracted from four dimensions: development intensity, heritage elements, river elements and vegetation elements, and street spatial scale feature indexes are constructed to characterize the spatial environment features. The model building and prediction correction module is used to build a visual responsibility deviation prediction model using machine learning methods and correct the prediction results. It takes visual responsibility deviation data and spatial scale feature index data generated by the input data acquisition module and the feature index extraction module as input, uses machine learning methods to build a visual responsibility deviation prediction model, predicts and evaluates the visual responsibility deviation of linear cultural heritage, and corrects the computer simulation visual responsibility measurement results based on the predicted deviation results, so that the corrected results are close to the actual perceived visual responsibility measurement results, thereby obtaining the optimized visual responsibility measurement results of linear cultural heritage and completing the visualization process.

Images