A night scene light environment AVC index evaluation method based on semantic segmentation

Abstract

The application provides a night scene light environment AVC index evaluation method based on semantic segmentation. The method comprises the following steps: step 1, using semantic segmentation technology to extract night scene visual information, light environment physical information and evaluation indexes of three dimensions of pedestrian attraction, vitality and carrying capacity, and constructing a multi-source data set; step 2, using statistical difference test and correlation analysis to identify and determine the night scene light environment elements affecting the three-dimensional evaluation indexes of AVC; step 3, using a nonlinear regression model and an interpretable analysis to obtain the nonlinear combination influence of different variable combinations on the three-dimensional evaluation indexes of AVC, and proposing a night scene light environment design optimization strategy. The application solves the problems of single dimension and insufficient fusion of subjective and objective data in the traditional night scene evaluation method, and realizes efficient, scientific and quantifiable night scene light environment evaluation.

Description

Technical Field

[0001] This invention relates to the field of urban lighting planning and built environment evaluation technology, and in particular to a method for evaluating the nighttime light environment AVC index based on semantic segmentation. Background Technology

[0002] With the development of the nighttime economy, urban nightscape lighting has shifted from purely functional lighting to a dual expression of culture and commerce. This is especially true for historical and cultural districts like Harbin's Central Street, where the lighting environment must simultaneously consider the attraction of historical features, the vitality of the commercial atmosphere, and the capacity of basic lighting. However, existing standards primarily focus on the compliance evaluation of physical indicators such as illuminance and brightness, lacking a comprehensive quantitative evaluation system that can coordinate aesthetic, commercial, and safety needs. In actual renovations, two contradictions often arise: either excessive commercialization destroys the historical atmosphere, or an overemphasis on preservation leads to insufficient commercial vitality.

[0003] Existing night scene evaluation methods primarily rely on physical parameter measurements and subjective questionnaire scoring. Physical indicators (such as brightness) cannot distinguish whether the light source is a historical building, an exquisite relief, or a cluttered commercial signboard, leading to drastically different visual and psychological experiences even with the same brightness data. Traditional methods lack refined quantification of spatial elements within the field of vision (such as buildings, signs, crowds, and vegetation), making it difficult to accurately calculate the specific proportion of each semantic object in the field of vision. This often results in evaluation results that are merely qualitative descriptions, lacking support from objective data.

[0004] At the data analysis level, existing research mostly employs traditional statistical methods such as linear regression to explore the relationship between lighting environment elements and subjective evaluations. However, human perception of the lighting environment exhibits significant nonlinear characteristics, making it difficult for traditional models to capture this complex nonlinear interaction. Therefore, there is an urgent need for an intelligent evaluation method that combines computer vision and machine learning algorithms to transform abstract visual perception into calculable pixel data, thereby achieving accurate prediction and scientific guidance for the quality of nighttime lighting environments. Summary of the Invention

[0005] This invention addresses the problems in existing night scene light environment evaluation, such as the disconnect between physical indicators and visual perception, the lack of quantitative means for "spatial semantic elements," and the inability of traditional linear analysis to reveal the interaction mechanism of complex elements. It proposes a night scene light environment AVC index evaluation method based on semantic segmentation.

[0006] This invention is achieved through the following technical solution: This invention proposes a method for evaluating the AVC index of night scene light environment based on semantic segmentation, the method comprising: Step 1: In the selected typical historical and cultural blocks, collect night scene image data, physical light environment parameters and AVC subjective evaluation data to construct a comprehensive evaluation dataset of night scene light environment; Step 2: To investigate the linear impact of semantic elements on night scene evaluation, statistical difference tests and correlation analysis are used to screen features in the comprehensive evaluation dataset to obtain independent linear association results for different features. Step 3: To address the nonlinear influence and interaction of multiple spatial elements obtained from semantic segmentation on night scene evaluation, a tree model based on machine learning is used to perform feature analysis on the comprehensive evaluation dataset. The interpretability analysis is then combined with the SHAP method to further derive the influence mechanism and establish a night scene light environment optimization strategy.

[0007] Furthermore, step 1 includes: determining the experimental site, selecting a typical pedestrian street with high commercial activity and historical features, and statistically analyzing its spatial node types.

[0008] Further, step 1 includes: acquiring semantic image data, taking multi-view night scene photos, and using the Mask2Former semantic segmentation model to extract five types of semantic variables: the proportion of historical buildings, the proportion of commercial activity, the proportion of road space, the population density, and the light environment coverage.

[0009] Furthermore, step 1 includes: acquiring objective physical data, and using a lux meter and a spectrometer to measure the road surface illuminance, color temperature and color rendering index on-site.

[0010] Furthermore, step 1 includes: obtaining subjective evaluation data, designing a scale based on AVC theory, and collecting public satisfaction ratings for the night view through a questionnaire survey.

[0011] Furthermore, the AVC theory refers to Attraction, Vitality, and Capacity; Attraction: focuses on the aesthetic value of the landscape, including the artistic feel of architectural lighting, the clarity of the outline, and the recognizability of the night view; Vitality: focuses on the intensity of activities, including the liveliness of the commercial atmosphere, the attractiveness of shops, and the tendency to stay; Capacity: focuses on the support of the environment, including pedestrian safety, visual comfort, and psychological warmth.

[0012] Further, step 2 includes: Step 2.1: Use Cronbach's alpha coefficient to test the reliability of the questionnaire, and use KMO value and Bartlett's test of sphericity to verify the validity of the data, so as to ensure the reliability of the subjective evaluation data; Step 2.2: Perform independent samples t-tests or ANOVA on the data of people of different genders, ages and identities to determine the significance of population attributes on AVC scores. Step 2.3: Construct the Pearson correlation coefficient matrix, analyze the linear correlation strength between each semantic independent variable and the AVC dependent variable, and use a heatmap visualization to screen out significantly correlated variables, eliminate multicollinear redundant variables, and determine the effective feature set for subsequent models.

[0013] Further, step 3 includes: Step 3.1: Standardize the data to meet the learning conditions of the tree model algorithm; Step 3.2: Use random forest to construct a nonlinear regression model and train the semantic independent variables to predict the AVC score; Step 3.3: Based on the ranking of the importance of the output features from the optimal model, identify the core driving factors that affect “attractiveness”, “vitality”, and “carrying capacity”. Step 3.4: Use the SHAP method to perform interpretability analysis on the optimal model, quantifying the marginal contribution of specific elements to the attractiveness and vitality of the night scene and the nonlinear threshold effect.

[0014] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the semantic segmentation-based night scene light environment AVC index evaluation method.

[0015] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the semantic segmentation-based night scene light environment AVC index evaluation method.

[0016] The beneficial effects of this invention are: This invention utilizes semantic segmentation technology to achieve pixel-level precision quantification of abstract visual concepts such as historical features and commercial interference, solving the problem that traditional physical indicators cannot characterize the content attributes of the light environment. Based on AVC triad theory and machine learning threshold analysis, it can not only evaluate the brightness and darkness of night scenes but also reveal the nonlinear relationship between commercial vitality and historical preservation, providing a scientific quantitative threshold reference for the refined management of night scenes in historical districts. The method described in this invention integrates subjective perception difference testing with nonlinear regression of objective data, avoiding the arbitrariness of single subjective scoring and the one-sidedness of single physical measurement, and establishing a replicable and scalable comprehensive evaluation technology system for urban night scenes. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 This is a flowchart of a night scene light environment AVC index evaluation method based on semantic segmentation according to the present invention.

[0019] Figure 2 This is a schematic diagram illustrating the semantic segmentation effect of night scene images based on the Mask2Former model in the case study.

[0020] Figure 3 This is a tree diagram illustrating the automated determination logic for the night scene type in the case study.

[0021] Figure 4 The feature importance ranking diagram based on the random forest regression model in the case study shows the influence weights of different semantic and physical variables on the night scene evaluation.

[0022] Figure 5 The SHAP dependency analysis plot for the key semantic variable in the case (taking the proportion of business vitality as an example) shows the nonlinear threshold effect of this variable on the evaluation results. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Specifically, in combination Figures 1-5 This invention proposes a method for evaluating the AVC index of night scene light environment based on semantic segmentation, the method comprising: Step 1: In the selected typical historical and cultural blocks, collect night scene image data, physical light environment parameters and AVC subjective evaluation data to construct a comprehensive evaluation dataset of night scene light environment; Step 2: To investigate the linear impact of semantic elements on night scene evaluation, statistical difference tests and correlation analysis are used to screen features in the comprehensive evaluation dataset to obtain independent linear association results for different features; key physical parameters with significant linear impact on AVC evaluation are screened out through correlation analysis. Step 3: Addressing the nonlinear impact and interaction of spatial multiple elements obtained from semantic segmentation on night scene evaluation, a tree model based on machine learning is used to perform feature analysis on the comprehensive evaluation dataset. Interpretability analysis is then conducted using the SHAP method to further derive the influencing mechanism and establish a night scene light environment optimization strategy. Specifically: a predictive model from objective variables to subjective AVC scores is constructed using machine learning algorithms to explore the nonlinear mapping relationships between variables. Interpretability analysis methods are introduced to quantify the marginal contribution and threshold effect of each element.

[0025] Furthermore, step 1 includes: determining the experimental site, selecting typical pedestrian streets with high commercial activity and historical features, and statistically analyzing their spatial node types (such as plaza nodes, festival decorations, historical buildings, and commercial districts).

[0026] Further, step 1 includes: acquiring semantic image data, taking multi-view night scene photos, and using the Mask2Former (Masked-attention Mask Transformer) semantic segmentation model to extract five types of semantic variables: the proportion of historical buildings, the proportion of commercial activity, the proportion of road space, the population density, and the light environment coverage.

[0027] Furthermore, step 1 includes: acquiring objective physical data, and using a lux meter and a spectrometer to measure the road surface illuminance, color temperature and color rendering index on-site.

[0028] Furthermore, step 1 includes: obtaining subjective evaluation data, designing a scale based on AVC theory, and collecting public satisfaction ratings for the night view through a questionnaire survey.

[0029] Furthermore, the AVC theory refers to Attraction, Vitality, and Capacity; Attraction: focuses on the aesthetic value of the landscape, including the artistic feel of architectural lighting, the clarity of the outline, and the recognizability of the night view; Vitality: focuses on the intensity of activities, including the liveliness of the commercial atmosphere, the attractiveness of shops, and the tendency to stay; Capacity: focuses on the support of the environment, including pedestrian safety, visual comfort, and psychological warmth.

[0030] Further, step 2 includes: Step 2.1: Use Cronbach's alpha coefficient to test the reliability of the questionnaire, and use KMO value and Bartlett's test of sphericity to verify the validity of the data, so as to ensure the reliability of the subjective evaluation data; Step 2.2: Perform independent samples t-tests or ANOVA on the data of people of different genders, ages and identities to determine the significance of population attributes on AVC scores and eliminate systematic bias caused by population background. Step 2.3: Construct the Pearson correlation coefficient matrix, analyze the linear correlation strength between each semantic independent variable and the AVC dependent variable, and use a heatmap visualization to screen out significantly correlated variables, eliminate multicollinear redundant variables, and determine the effective feature set for subsequent models.

[0031] Further, step 3 includes: Step 3.1: Standardize the data to meet the learning conditions of the tree model algorithm; specifically, use standardization and regularization to preprocess the data, reduce the contribution of feature amplitude, and reduce the overfitting rate. Step 3.2: Construct a nonlinear regression model using Random Forest to train the semantic independent variables' predictive ability for AVC scores; specifically, train the model on the data using Random Forest and utilize R... 2 Indicators such as RMSE are used to evaluate the fit. The optimal nonlinear regression model is obtained through testing and parameter tuning. Step 3.3: Based on the ranking of feature importance output by the optimal model, identify the core driving factors affecting "attractiveness", "vitality" and "carrying capacity" (such as the contribution weight of the proportion of business vitality to vitality). Step 3.4: Use the SHAP (Shapley Additive exPlanations) method to perform interpretability analysis on the optimal model, quantify the marginal contribution of specific elements to the attractiveness and vitality of the night view and the nonlinear threshold effect. Specifically, introduce the SHAP (SHapley Additive exPlanations) method, draw SHAP summary diagrams and dependency diagrams, intuitively show the positive and negative impacts of specific semantic elements (such as the proportion of historical buildings) on the quality of the night view in different value ranges, and identify the key thresholds for maintaining the overall balance of AVC.

[0032] This invention proposes a semantic segmentation-based evaluation method for nighttime light environment attraction, viability of commercial atmosphere, and capacity of basic lighting, referred to as AVC index, belonging to the field of smart city and architectural lighting environment technology. The method includes the following steps: Step 1, using semantic segmentation technology to extract nighttime visual information, physical information of the lighting environment, and evaluation indicators for three dimensions: pedestrian attraction, viability, and capacity (AVC), constructing a multi-source dataset; Step 2, using statistical difference tests and correlation analysis to identify and determine the nighttime light environment elements affecting the three dimensions of AVC evaluation indicators; Step 3, using nonlinear regression models and interpretable analysis to obtain the nonlinear combined influence of different variable combinations on the three dimensions of AVC evaluation indicators, and proposing nighttime light environment design optimization strategies. This invention solves the problems of single dimensions and insufficient integration of subjective and objective data in traditional nighttime evaluation methods, achieving efficient, scientific, and quantifiable evaluation of the nighttime light environment.

[0033] Example Combination Figure 1 This embodiment of the invention proposes a semantic segmentation-based AVC index evaluation method for nighttime light environments. Its core lies in constructing a comprehensive evaluation system for the visual and psychological effects of nighttime light environments on pedestrians, used for nighttime light environment optimization. Specifically, it includes the following three implementation steps: Step 1: In the selected typical historical and cultural blocks, collect night scene image data from human viewpoint, physical light environment parameters, and subjective evaluation data based on AVC theory simultaneously to construct a multi-dimensional comprehensive evaluation dataset of night scene light environment. Step 2: Through correlation analysis, key physical parameters with significant linear impact on AVC evaluation are screened out, and parameters with independent linear impact on night scene perception are screened out. Step 3: Use the random forest tree model under machine learning to perform feature analysis on the dataset, and combine it with the SHAP interpretability method to analyze the nonlinear influence and interaction of non-light environment and light environment elements such as space, gender, and age on night scene evaluation. Further, step 1 specifically includes: Step 1.1: Select a typical pedestrian street with both "high commercial activity" and "rich historical character" as the experimental subject. Sampling points were set up at intervals of 50-100 meters along the main street and core side streets. An automated scene classification logic was established based on spatial semantic features to avoid the subjective bias of manual classification. For example... Figure 2As shown, the experimental scenes were scientifically divided into four categories: festive decoration landscapes, plaza node landscapes, historical building landscapes, and commercial street landscapes. The criteria for classifying these four categories were as follows: if the semantic pixel proportion of "festive installations (such as ice sculptures and light galleries)" in the image exceeded 10%, it was prioritized as a festive decoration landscape. This type of scene has extremely high instantaneous visual impact and is a special form of winter nightscape in cold regions; if the pixel proportion of "road surfaces and plaza paving" in the image exceeded 40%, and there was no significant continuous commercial interface in the field of view (commercial proportion <15%), it was judged as a plaza node, which mainly serves the functions of gathering and dispersing traffic; for the remaining scenes, the classification was determined by comparing the pixel proportions of historical building facades and commercial activity areas. If the proportion of historical buildings was higher than that of commercial areas, it was judged as a historical building landscape, focusing on the nighttime display of the building texture; otherwise, it was judged as a commercial street landscape, focusing on the creation of a commercial atmosphere.

[0034] Step 1.2: Acquire objective physical data and high-precision semantic image data. At the same location and the same line of sight (1.6m) where the photo was taken, use professional optical instruments to collect physical data synchronously. Use a lux meter to measure the illuminance of the horizontal road surface to reflect the basic functional lighting level. The correlated color temperature and color rendering index of the light source reflect the warm and cool properties of the light color and the ability to reproduce the color of objects. Step 1.3 involves simultaneous semantic data extraction. The Mask2Former model is used to perform semantic segmentation on the 460 collected night scene photos, such as... Figure 2 As shown. Addressing the technical challenges in night scene images, such as the blurring of dark areas with the night sky and the obscuring of building outlines by signage halos, and the problem of indistinct boundaries between commercial signs and building facades and road surfaces and dark areas under complex nighttime lighting conditions, this study collects daytime images from the same viewpoint for auxiliary segmentation. The clear semantic boundaries from the daytime images are used to correct the nighttime semantic mask, thereby accurately extracting the following semantic variables: Historical building proportion: the pixel proportion of pure historical building facades after removing interference from commercial signs and light boxes; Commercial activity proportion: the pixel proportion of high-brightness areas (including signs, shop windows, and electronic screens) in the field of view; Road space proportion: the pixel proportion of ground areas accessible to pedestrians; Crowd density: the pixel proportion of pedestrians and tourists in the field of view; Light environment coverage: a reverse indicator of the proportion of sky and unlit dark areas in the field of view.

[0035] Step 1.4: Obtain subjective evaluation data and design a structured questionnaire based on the AVC triad theory (Attraction, Vitality, and Capacity) in environmental psychology. Attraction (A): Focuses on the aesthetic value of the landscape, including the artistic appeal of architectural lighting, the clarity of the outline, and the recognizability of the night view; Vitality (V): Focuses on the intensity of activity, including the liveliness of the commercial atmosphere, the attractiveness of shops, and the tendency to linger; Capacity (C): Focuses on the supportive capacity of the environment, including pedestrian safety, visual comfort (no glare), and psychological warmth.

[0036] Furthermore, step 2 specifically includes: Step 2.1: Ensure data validity by conducting reliability and validity analyses on the questionnaire data. Reliability analysis used Cronbach's alpha, and validity analysis used KMO and Bartlett's test of sphericity. All dimensions showed Cronbach's alpha values ​​greater than 0.8 and KMO values ​​greater than 0.7 with a significance P < 0.001, indicating that the data possesses extremely high internal consistency and construct validity, and can be used for subsequent modeling analysis. Step 2.2 involves identifying potential interference from user attributes on the evaluation results and conducting a difference analysis. Independent samples t-tests (for gender) and one-way ANOVA (for age and identity) were used. The analysis showed a statistically significant difference in the "Carry-Through (C)" dimension scores between genders (P=0.032<0.05), specifically, female subjects showed significantly higher sensitivity to nighttime darkness and glare than male subjects. Therefore, the impact weight of safety lighting on female users should be fully considered in the subsequent construction of the evaluation model and strategy formulation.

[0037] Step 2.3: Remove linearly uncorrelated or redundant variables and construct the Pearson correlation coefficient matrix. To avoid multicollinearity leading to distorted model parameter estimation, this embodiment follows the "semantic priority" principle, retaining semantic variables that have stronger explanatory power for AVC and contain more information, and determining the effective feature set for step 3 to meet the learning conditions of the machine learning algorithm.

[0038] Furthermore, step 3 specifically includes: Step 3.1: Satisfy the training conditions for statistical machine learning, eliminate the numerical differences between different units, such as the pixel percentage % and the color temperature K value, and perform Z-score standardization on all continuous variables to make the data satisfy the normal distribution assumption while reducing the impact of feature amplitude on prediction accuracy and improving the model's generalization ability. Step 3.2: Construct the nonlinear mapping relationship using the Random Forest Regression algorithm. Random Forest is an ensemble learning algorithm based on the Bagging strategy. It makes predictions by constructing multiple decision trees and taking their average, effectively handling high-dimensional data and capturing nonlinear relationships and threshold effects between variables. To verify the model performance, five-fold cross-validation is used to calculate the coefficient of determination (R²). 2 ) and root mean square error (RMSE); Step 3.3: Obtain the non-linear impact weights of each element on night scene perception, and combine them with the optimal random forest model to output the feature importance ranking. For example... Figure 4 As shown, the model quantifies the contribution of each element and clarifies the core driving variables of different evaluation dimensions.

[0039] Step 3.4 further analyzes the interaction relationships and specific thresholds between elements using the Shapley Additive Explanations (SHAP) method. SHAP values, based on game theory principles, decompose the model prediction results into the marginal contribution values ​​of each feature. For example... Figure 5 The SHAP dependency analysis graph is used. Based on the threshold and feature importance ranking determined by SHAP analysis, a refined optimization strategy for hierarchical classification is formulated.

[0040] In summary, this invention, by introducing deep learning semantic segmentation technology, achieves pixel-level quantification of abstract concepts such as "historical features" and "commercial atmosphere"; through machine learning and SHAP interpretability analysis, it not only identifies key influencing variables but also quantifies the optimal range and critical threshold of each variable, providing scientific and objective data support for the refined lighting planning and management of historical and cultural blocks.

[0041] The present invention also proposes an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the semantic segmentation-based night scene light environment AVC index evaluation method.

[0042] The present invention also proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the semantic segmentation-based night scene light environment AVC index evaluation method.

[0043] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0044] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0045] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0046] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as execution by a hardware decoding processor, or as a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0047] The above provides a detailed description of the AVC index evaluation method for night scene light environment based on semantic segmentation proposed in this invention. Specific examples have been used to illustrate the principle and implementation of this invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for evaluating the AVC index of night scene light environment based on semantic segmentation, characterized in that, The method includes: Step 1: In the selected typical historical and cultural blocks, collect night scene image data, physical light environment parameters and AVC subjective evaluation data to construct a comprehensive evaluation dataset of night scene light environment; Step 2: To investigate the linear impact of semantic elements on night scene evaluation, statistical difference tests and correlation analysis are used to screen features in the comprehensive evaluation dataset to obtain independent linear association results for different features. Step 3: For the nonlinear influence and interaction of spatial multi-factors obtained by semantic segmentation on night scene evaluation, the tree model under machine learning is used to perform feature analysis on the comprehensive evaluation dataset, and the interpretability analysis is performed by combining the SHAP method to further derive the influence mechanism and establish a night scene light environment optimization strategy. Step 1 includes: acquiring semantic image data, taking multi-viewpoint night scene photos, and using the Mask2Former semantic segmentation model to collect daytime images from the same viewpoint for auxiliary segmentation, using clear daytime semantic boundaries to correct the nighttime semantic mask, and extracting five types of semantic variables: the proportion of historical buildings, the proportion of commercial activity, the proportion of road space, the population density, and the light environment coverage. Step 2 includes: Step 2.1: Use Cronbach's alpha coefficient to test the reliability of the questionnaire, and use KMO value and Bartlett's test of sphericity to verify the validity of the data, so as to ensure the reliability of the subjective evaluation data; Step 2.2: Perform independent samples t-tests or ANOVA on the data of people of different genders, ages and identities to determine the significance of population attributes on AVC scores. Step 2.3: Construct the Pearson correlation coefficient matrix, analyze the linear correlation strength between each semantic independent variable and the AVC dependent variable, and use a heatmap visualization to screen out significantly correlated variables, eliminate multicollinear redundant variables, and determine the effective feature set for subsequent models.

2. The method according to claim 1, characterized in that, Step 1 includes: determining the experimental site, selecting a typical pedestrian street with high commercial activity and historical features, and statistically analyzing its spatial node types.

3. The method according to claim 1, characterized in that, Step 1 includes: acquiring objective physical data, and using a lux meter and a spectrometer to measure the road surface illuminance, color temperature and color rendering index on the spot.

4. The method according to claim 1, characterized in that, Step 1 includes: obtaining subjective evaluation data, designing a scale based on AVC theory, and collecting public satisfaction ratings for the night view through a questionnaire survey.

5. The method according to claim 4, characterized in that, The AVC theory refers to Attraction, Vitality, and Capacity. Attractiveness: Focus on the aesthetic value of the landscape, including the artistic feel of the architectural lighting, the clarity of the outline, and the visibility of the night view; Vitality: Focus on the intensity of the activity, including the liveliness of the commercial atmosphere, the attractiveness of the stores, and the tendency for people to linger; Bearing capacity: Focus on the support capacity of the environment, including walking safety, visual comfort and psychological warmth.

6. The method according to claim 1, characterized in that, Step 3 includes: Step 3.1: Standardize the data to meet the learning conditions of the tree model algorithm; Step 3.2: Use random forest to construct a nonlinear regression model and train the semantic independent variables to predict the AVC score; Step 3.3: Based on the ranking of the importance of the output features from the optimal model, identify the core driving factors affecting "attractiveness", "vitality", and "carrying capacity"; Step 3.4: Use the SHAP method to perform interpretability analysis on the optimal model, quantifying the marginal contribution of specific elements to the attractiveness and vitality of the night scene and the nonlinear threshold effect.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-6.

8. A computer-readable storage medium for storing computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-6.

Images