Comparative Analysis of TAG (Three-Dimensional Architectural Greening) Scenic Beauty Quantitative Techniques Based on Visual Perception

Abstract

Three-dimensional architectural greening (TAG) enables the integration of ecological, economic, and social advantages via the effective use of multidimensional space in a variety of forms, making it a significant method for enhancing spatial quality in densely populated cities. TAG technology has expanded the scope and capabilities of urban greening. It has the ability to provide green space, improve urban ecology and aesthetics, and alleviate the conflict between limited land resources and rising demand for greening throughout the urbanization process. Currently, there is a lack of a systematic assessment approach that focuses on the public’s visual perception of TAG. It is critical to focus on advances in visual perception approaches and create a “people-oriented perception driven” evaluation system that serves as a scientific foundation for urban three-dimensional greening initiatives. First, this study created a database of 300 TAG cases and selected classic cases using screening, classification, and sampling. Second, three experiments were set up for the study, including the use of the semantic differential (SD) method, and scenic beauty estimation (SBE) for subjective evaluation, and the eye-tracking experiment for objective evaluation. Finally, this study compared subjective and objective evaluations and demonstrated that both two approaches had a certain amount of accuracy. It also investigated the relationship between spatial features and public visual perceptions using methods such as factor and correlation analysis. The three effective methods for evaluating the quality of TAG based on visual perception that are presented in this study—two subjective and one objective—use standardized images, are quick and simple to use, and make up for the drawbacks of conventional strategies like indirectness, inefficiency, and time-consuming data collection. They also form a solid foundation for the real-world application of categorization prediction. In addition to being adaptable to a wide range of application settings, these two assessment paths—subjective evaluation and objective evaluation—can be integrated to complement one another and provide scientific references for future TAG designs and spatial decision making.

1. Introduction

Land resources in large cities and urban core areas have become increasingly scarce since the beginning of the 21st century, as China’s urbanization has accelerated. With a supplement to traditional horizontal greening, three-dimensional urban greening plays an essential role in constructing Healthy Cities and Garden Cities [1,2]. Three-dimensional architectural greening (TAG) extends typical flat greening into a three-dimensional form by covering building facades including walls, roofs, and balconies with plants, considerably reducing the demand for ground land resources. Greening efficiency per unit area can be up to 20 times higher than regular greening. In recent years, TAG technologies have become a widely applicable type of urban greening, covering various dimensions of buildings such as roofs, balconies, and facades [3]. By integrating architecture and greening design, it achieves the ecological and aesthetic fusion of space [4,5]. However, the aesthetic value of TAG is frequently disregarded [6], and its visual impact on urban image and spatial experience has not yet been properly quantified [7,8]. In light of this, assessing how urban dwellers perceive TAG visually and establishing a connection between evaluation research and design guidelines have become crucial areas of study in this area [9,10]. Analyzing the key design elements that influence public visual perception might provide strategic references for future TAG designs. This study fully incorporates visual perception principles and presents evaluation paths for the quality of TAG via subjective aesthetic experience and empirical data analysis. The goal is to provide a highly scientific and forward-thinking reference for creative three-dimensional greening design and intelligent decision making in urban spatial layouts, as well as to encourage the harmonious cohabitation of green buildings and urban aesthetics.

Modern urban facilities consume a substantial quantity of land resources, and the conflict between supply and demand of urban land resources is becoming exacerbated against the backdrop of accelerated new urbanization. The average plot ratio of urban built-up areas has risen to over 2.8 due to intensive development strategies like smart city construction and old community renovation, which directly compresses green space, according to the Ministry of Housing and Urban Rural Development’s 2022 urban renewal pilot project in 21 Chinese cities [11]. As traditional flat greening becomes increasingly limited, TAG has steadily evolved as a viable alternative and an important component of urban environments in the new era [12,13]. Currently, there is a strong academic base for evaluating landscape visual quality, with an emphasis on sustainability and ecological benefits [14,15]. In recent years, the focus has switched to the social and cultural implications. Nevzati F. et al. developed an evaluation system by combining landscape features evaluation and the cultural ecosystem services framework, evaluating the landscape value of the suburban area of Harku, Estonia [16]. Nelson R. undertook a landscape evaluation of the Rabigh Coast region in Saudi Arabia, offering a thorough factor score to aid decision making [17]. Luo Y. et al. employed subjective assessment data for quantitative analysis and offered recommendations for enhancements for actual cases [18]. It is widely accepted in academia that the psychophysics of landscape visual evaluation seen to be more trustworthy in related study cognition [19]. The scenic beauty estimation (SBE) approach and the semantic differential (SD) method are the two research techniques most commonly used in studies.

Although subjective landscape visual evaluation can partially quantify public opinion, practical applications are prone to prejudice [20]. Over the past few years, new technology gadgets such as Geographic Information Systems (GIS), virtual reality (VR) devices, and eye-tracking instruments have made significant progress in the field of landscape visual evaluation [21,22]. Among these, eye-tracking technology has a solid research foundation. Initially, it was widely used in disciplines such as medical, education, transportation, and human–computer interaction [23,24], but it has been gradually incorporated into geography, tourism, and landscape [25,26,27]. Eye-tracking technologies can gather and evaluate data on human behavior, revealing the underlying link between visual attention and spatial properties [28]. In architecture and landscape projects, eye-tracking techniques have significantly improved landscape treatment, perception, and evaluation [29,30,31]. Amati M. et al. discovered that under stressful conditions, participants spent more time gazing at trees and shrubs [32]. Półrolniczak M. et al. discovered through eye-tracking experiments that participants focused more on information from the landscape under negative weather conditions than in positive weather conditions [33]. Misthos L-M. et al. observed that an increase in the visual size of a quarry led to a reduction in participants’ visual attention [34].

Although research employing subjective psychological evaluation or physiological information gathering methods has achieved some success in the disciplines of architecture and landscape currently, there are still some limitations. First, there are still constraints in assessing accuracy and collecting data information [35,36]. The scoring results in subjective evaluation are greatly impacted by intrinsic elements, including the evaluators’ cultural background and aesthetic experience [37]. Individual subjective preference interference is still a challenge in the scientific measurement of landscape visual resource evaluation indicators like field of view range and perspective selection [38]. The popularity of visual perception experiments is limited by issues with eye-tracking equipment, such as their expensive cost and complicated operation [39]. According to research, devices must synchronously collect multidimensional data, including changes in pupil diameter (with an accuracy of 0.1 mm) and gaze duration (with an error of ±50 ms), in perception experiments conducted in dynamic visual environments. This places a tremendous amount of strain on the experimental environment control and data processing capabilities [40]. It is clear that the evaluation of TAG urgently has to concentrate on innovations in visual perception-based methodologies, and develop a “human centered perception driven” evaluation system. Second, existing studies commonly use a single method: either subjective psychological evaluation or physiological information gathering. The limitations of each method may compromise the scientificity of results. Therefore, the evaluation of TAG urgently requires integrating subjective psychological evaluation with objective physiological data. Cross-validating two types of results will establish a comprehensive perception evaluation system, providing scientific support for designing people-centered urban landscapes.

2. Methods

The physical surroundings and public visual perception, which include both objective and subjective elements, have an impact on how TAG scenic beauty is perceived. Therefore, in order to explore the TAG scenic beauty based on visual perception, three stages were set up in order to investigate the visual perception-based evaluation path and confirm its logicalness and scientificity: the experimental preparation, evaluation experiment, and the comparative analysis (Figure 1). For more information of key resource, see to Appendix A.

• Experimental preparation: To provide representative examples for the visual perception-based scenic beauty study, a TAG sample case library was first created. The experimental preparation arrangement includes sample screening (158 samples), participant screening (37 people), and venue and equipment (eye-tracking laboratory).
• Evaluation experiment: Three experiments were performed to assess the scenic beauty of typical samples, using the SBE approach, SD method, and eye-tracking method. Experiment Ⅰ and Experiment Ⅱ were subjective, while Experiment Ⅲ was objective.
• Comparative analysis: The correlation between subjective and objective evaluations was studied using comparison analysis, which not only compares the three assessing formulas of scenic beauty proposed in this study but also verifies their scientific validity.

2.1. TAG Case Library

The implementation of the TAG project is influenced by a variety of circumstances, including national conditions, regions, the economy, and so on, with considerable regional disparities [41]. In order to ensure the scientificity of the evaluation, this study first must prepare a case library with a certain coverage breadth. Therefore, in the stage of experimental preparation, 536 TAG cases were collected for the study using a variety of techniques, including literature extraction, network data crawling, and field investigation. A preliminary screening was then carried out to obtain 482 cases, which were archived in the form of images, in accordance with the following principles: diversity (covering various forms and technologies of cases), representativeness (typical projects), scientificity (accurate data), ecological priority (ecological benefits), cultural integration (local characteristics), operability (maintenance costs, etc.), and safety standards (structural safety). Finally, an experimental case library was built using 300 cases, taking into account the comprehensiveness and balance of technological and geographical types. Each case includes high-quality architectural photographs from multiple angles for experimental selection. These images allow the selection of photos with consistent visual effects, which feature uniform lighting, similar composition, and mature vegetation. Rendered images, low-quality photographs, and edited images were minimized to reduce variations among sample images and ensure the scientific validity of perception assessments.

These cases were categorized into two groups based on the technical groups utilized, and three types based on spatial composition properties (Figure 2). The data covered all spatial subtypes. Following the initial screening, 158 samples (158 images) were selected for Experiment Ⅰ, 40 samples (40 images) were selected for Experiments Ⅱ and Ⅲ-1, and 12 samples (12 images) were selected for Experiment Ⅲ-2. Every subtype of spatial classification is represented in the sample set chosen for each experiment.

There are currently still divergent views in scholarly circles over how to categorize three-dimensional greening, because different studies offer distinct categorization frameworks based on function, technical form, or spatial dimensions. Countries like Germany promote classification standardization through rules and regulations (such as eco-performance-oriented technology grading), but China calls for the implementation of a “function–technology–space” tripartite classification framework in practice. In the visual perceptual interaction experiments set up in this study, the formation of all visual impressions is positively connected with spatial form and is influenced by the requisite technologies. As a result, in the establishment of the case library and the selection of representative samples, spatial classification takes precedence, with technical classification serving as an important reference.

2.1.1. Technical Types

TAG can be divided into Green Roofs and Green Facades based on technical types (

Table 1. Representative photographs of technical types.

Technical Type	Representative Photographs
Green Roofs Project
	Green Hill, Shanghai Building Complex (Original Tobacco Company Machine Repair Warehouse)	Nanjing Tech University, Nanjing, Educational Buildings
Green Facades Name
	Tongji University, Shanghai, Educational Buildings	Cangjie Commercial Plaza, Suzhou, Open Business Center (Original Jiangsu Third Prison)

). With this scientific background, Indoor Greening can be viewed as a hybrid of Green Roof and Green Façade technologies in indoor spaces.

Green Roofs typically involve planting vegetation on the building’s roof. According to the form, species, and planting techniques of the plants, Green Roofs can be classed into three types based on their plant form, species, and planting techniques: extensive, semi-intensive, and intensive [42]. Among these three types, the plant species range from grasses to shrubs to trees, with the variety of plant species increasing and the greening volume gradually increasing as well. Green Façades are the planting of plants on vertical walls. They are classified into three varieties based on their planting techniques: climbing and trailing greening systems, planter box systems, and modular green wall systems. Climbing and trailing greening systems make use of climbing plants, such as vines, that attach to walls and grow upward. This can be classified into three subcategories: climbing greening without support, wall-supported climbing greening, and three-dimensional grid systems [43]. Planter box systems refer to plants that are cultivated in individual spaces, where the containers can either be placed outside the wall or designed as wall-integrated planting boxes. Modular green wall systems involve wall modules pre-planted with vegetation that are assembled together to form a green wall [39].

2.1.2. Spatial Types

Based on the location characteristics of greening, TAG may also be divided into three major groups: Horizontal Greening Systems, Vertical Greening Systems, and Indoor Greening (

Table 2. Representative photographs of spatial types.

Spatial Type	Representative Photographs
Horizontal Greening Systems
Name	SEOULLO 7017, South Korea, Pedestrian Street	Green Hill, Shanghai, Building Complex (Original Tobacco Company Machine Repair Warehouse)
Vertical Greening Systems
Name	HKRI Taikoo Hui, Shanghai, Commercial Complex	Sanke Yinghua of Cangjie Commercial Plaza, Suzhou, Open Business Center (Original Jiangsu Third Prison)
Indoor Greening
Name	The Roof, Shanghai, Commercial Complex	1000 TREES, Shanghai, Commercial, Hotel, and Office Buildings

). Each type is further classified according to the various forms of construction and greening space, resulting in a systematic classification of TAG spaces that encompasses all research situations. The experimental investigation in this study was based on the spatial types of TAG [44]. The data in Figure 1 depict the number of cases by kind. The following designations are used: “Type A” refers to Vertical Greening Systems, “Type B” refers to Horizontal Greening Systems, and “Type C” refers to Indoor Greening.

2.2. Experimental Preparation

2.2.1. Sample Screening

This study utilized the images of various cases as research material, necessitating graphic processing of the samples and selection of representative cases. It consisted of the following steps:

First, the ratio and precision of the images were harmonized, and for each case, the best representative image was chosen. The image needed to have adequate daytime illumination to fully capture the theme effect of TAG.

Second, the images were cropped to a uniform horizontal composition in Photoshop, ensuring a 4:3 ratio, with a resolution of 1920 × 1440 pixels and 300 dpi.

Finally, a total of 158 cases were selected, yielding 158 images.

The three experiments were progressively linked, with the images used being 158, 40, and 52, respectively (Figure 3). Experiment Ⅰ included 158 images that were used. Given that the number of images in Experiment Ⅱ should be kept to a minimum (typically between 10 and 45), 40 images were chosen. The images were randomly stratified after being classified by the “σ principle” [45,46,47,48,49] using SPSS, resulting in 40 images for Experiments Ⅱ and Ⅲ. During this process, three pre-test photos for Experiment Ⅲ were chosen at random from the 40 images. These pre-test images were excluded from the formal data analysis to ensure the trials’ scientific validity and effectiveness. After assessing the scenic beauty values of 158 images, the 12 images with the highest scenic beauty value for each spatial subtype were selected for use in Experiment Ⅲ-2.

2.2.2. Participant Screening

The number of participants in each experiment was selected by taking into account aspects such as the accuracy of the experimental results, the difficulty of participation, and the ratio of participants in the experiment. According to relevant research, the number of participants in the SBE and SD method experiments is often managed between 20 and 50 persons [50], and psychological studies generally consider trials with more than 30 participants as big sample trials with higher reliability [51]. As the SD approach requires a basic understanding of language to quantify landscape features, the participants in this study were primarily university students. For the experiments, a total of 37 volunteers were chosen. With a roughly 1:1 male to female ratio, 10 of them were between the ages of 20 and 22, while 27 were between the ages of 24 and 26. Participants in the eye-tracking experiment also had to meet the following requirements.

• Uncorrected visual acuity of 4.8 or higher without the use of glasses or contact lenses.
• No makeup (mascara) or colored contact lenses.
• Enough sleep and a sound mental condition.

2.2.3. Venue and Equipment

To reduce the impact of environmental influences on experimental results, three experiments were carried out in the same eye-tracking laboratory studio, with the same spatial environment used throughout each of them. The laboratory has two entrances without windows. During each of the three experiments, the doors were closed, and the indoor lighting was turned on to avoid interference from external light, sound, and other factors, ensuring that the experiment was carried out under the same external conditions and that the data obtained were reliable and consistent.

As shown in Figure 4, Experiments Ⅰ and Ⅱ are primarily concerned with recording the participant’s visual perception evaluations, with the Dell monitor in the laboratory serving as a sequential playback device. Experiment Ⅲ was carried out using a computer and external eye-tracking equipment. There were two Dell monitors, one with eye-tracking equipment (aSeePro) for participants, and one to control the experiment. The equipment’s sampling rate was 100 Hz, with an operational distance of 50–80 cm. The device monitors a head movement range of 30 cm (width) × 25 cm (height) × 65 cm (depth) and includes two EyeSensor modules for coordinated data collection. The main components were infrared high-speed imaging devices, data transmission lines, and power supply systems. The equipment interface entailed connecting the infrared imaging sensors, which were positioned beneath the display bezel, to the host computer via data cables, allowing real-time data gathering and transmission to dedicated analysis software, aSeeStudio (version V2.0). This program not only supplied quantitative representations of eye-tracking data, but it also supported visual analysis such as heatmap creation, gaze trajectory mapping, and AOI analysis.

2.3. Evaluation Experiment

The two components of the subjective evaluation experiment are Experiment Ⅰ, which is based on the design of the SBE method, and Experiment Ⅱ, which is mainly based on the SD method. In addition to serving as an integrated component of Experiment Ⅱ, the design experiment based on the SBE approach may independently give a quantitative assessment of scenic beauty.

The objective evaluation experiment utilized the aSeePro telemetry eye tracker, and the experimental stimuli were displayed on a Dell monitor with a resolution of 1920 × 1080 dpi. To better understand how the audience perceives TAG visual beauty, the experiment was separated into two halves.

2.3.1. Experiment Ⅰ

The SBE method employed in Experiment Ⅰ accounts for the evaluators’ intuitive impression and preference rating of the TAG, yielding results that are intuitive and easy to compare horizontally. It should be noted that this experiment is based on subjective evaluation and requires a large number of participants (professional/non-professional) to score photos or real scenes. Furthermore, this method comprehensively obtains the scenic beauty value based on the characteristics of the TAG itself and the cognition of the subjects, and the results are easily influenced by the evaluators’ aesthetic preferences. As a result, Experiment Ⅰ was designed with 158 cases picked from the case library based on image quality screening, and 37 individuals were invited to participate.

• Subjective evaluation: From 158 cases, one representative image was selected from each case for the scenic beauty perception experiment (158 images were selected in total), and 37 participants were invited to evaluate the scenic beauty value of each image. A Likert five-point scale was used to ask 37 people to rate the scenic beauty value of each sample, and the average of their assessments was used to determine the scenic beauty of each sample. As indicated in Formula (1), SBE_i denotes the scenic beauty value of the i-th sample, whereas n represents the number of participants (in Experiment Ⅰ, n = 37).

SBE_i = (SBE_i1 +SBE_i2 + … +SBE_in)/n

(1)

• Classification and sorting: Experiment Ⅱ and Experiment Ⅲ were prepared in part by this experiment. The μ − σ and μ + σ were used as border points to classify each sample type into low, medium, and high interest levels in accordance with the statistical “σ principle”. At least one case was chosen from each subcategory using random stratified sampling, which was carried out in SPSS based on the percentage of the three TAG categories in the case collection. A total of 40 case samples were obtained to participate in the further experiment (Experiment Ⅱ).
• The scenic beauty values of each sample, which represent its overall visual aesthetics, was recorded and compared to the outcomes of Experiment Ⅱ and Experiment Ⅲ.

2.3.2. Experiment Ⅱ

The SD method, a widely used psychophysical technique, may accurately measure subjective impressions using verbal scales. After the completion of Experiment Ⅲ, an SD-based scenic beauty perception test (Experiment Ⅱ) was performed on the 40 images selected in the second screening. The participants rated the details of the landscape for each case.

• Evaluation system: The SD method requires the selection of evaluation factors, which should be based on mature evaluation systems. However, current special research on TAG evaluation systems remains limited. It cannot provide sufficient support for selecting factors. Therefore, this study focuses on TAG as a composite of architecture and landscapes, referring to a mature landscape evaluation system. This study particularly emphasizes the factors related to architecture and spatial design, making the evaluation system more applicable to architectural design. It investigated academic articles on landscape appraisal from the last five years on the platforms of Web of Science and China National Knowledge Infrastructure (CNKI), eliminating articles from the same author or team. A total of 46 high-impact publications were obtained, and initial landscape evaluation variables were identified. The landscape evaluation factors were then filtered further using a frequency-based selection approach. To create the subjective evaluation system for Experiment Ⅱ, seven evaluation variables were discovered by rating, merging related items, and removing low-impact factors. The subjective evaluation system for Experiment Ⅱ covers three major landscape elements based on urban spatial components: Spatial Elements, Natural Elements, and Architectural Elements. It is further subdivided into seven subjective evaluation factors: Landscape Subjectivity, Scene Balance, Vegetation Layering, Landscape Coordination, Architectural Coordination, Functional Integration, and Color Coordination (

  Table 3. Subjective evaluation factors of Experiment Ⅱ.

  Landscape Elements	Subjective Evaluation Factors	Inspection Content
  Spatial Elements	Landscape Subjectivity	Whether the three-dimensional landscape scene highlights the target presented in the picture.
  	Scene Balance	The compositional structure and balance of the elements in the picture.
  Natural Elements	Vegetation Layering	The layering of vegetation in the landscape, such as trees, shrubs, and grass.
  	Landscape Coordination	The coordination between the three-dimensional greening in the picture and the surrounding landscape.
  Architectural Elements	Architectural Coordination	The integration of the greening in the picture with the building and the surrounding architecture.
  	Functional Integration	The degree of integration between the three-dimensional greening and the functional aspects of the building.
  	Color Coordination	The degree of harmony between the greening and the building’s color scheme.

  ).
• Random stratified sampling: According to the statistical classification results of Experiment Ⅰ, this study employed SPSS random stratified sampling to recruit 40 samples for Experiment Ⅱ, guaranteeing full coverage and balance of the three types during random sampling.
• Subjective evaluation: For the 40 selected samples, 37 participants were invited to rate the scenic beauty value of each sample. The evaluation was conducted using a Likert five-point scale. Factor analysis was used to examine the evaluation influencing factors, extract principal components, and derive subjective evaluation formulas. The subjective evaluation of scenic beauty (SBESi) of each sample was calculated.
• Comparation: The scenic beauty values of each sample were recorded and compared to the outcomes of Experiment Ⅰ and Experiment Ⅲ.

2.3.3. Experiment Ⅲ

(1)

Experiment Ⅲ-1

Experiment Ⅲ-1 is a single-image experiment. Referring to the results of Experiment Ⅰ, 40 images were extracted in layers to participate in this stage, and the selected images were consistent with the samples of Experiment Ⅱ.

Then, 37 individuals were invited to engage in this single-image experiment, where 40 images were exhibited on a single sheet and their eye movement data were recorded. Among them, 6 indicators were selected for analysis: Total Fixation Duration, Average Fixation Duration, First Fixation Duration, Fixation Count, Saccade Count, and Average Saccade Amplitude.

• Total Fixation Duration displays the Total Fixation Duration of all fixation points in the area. Longer Total Fixation Duration indicates a higher level of attention to the image, suggesting that the image contains more information.
• Average Fixation Duration is related to scene analysis, understanding, perception, and cognition, reflecting the degree of comprehension of the image.
• First Fixation Duration reflects the initial recognition of the image. The shorter the first fixation time, the higher the initial recognition.
• Fixation Count reflects how much attention the participants paid to each area, indicating their interest in the image.
• Saccade Count reflects the search activities of the participants’ eyes, showing how many times they searched for details in the image.
• Average Saccade Amplitude is the average value of all saccadic movement amplitudes. Saccade amplitude is the spatial range of eye movement from the original fixation point to the goal fixation point. A higher Average Saccade Amplitude, which is typically impacted by semantic information, implies a lower interest in the material between fixation sites.

(2)

Experiment Ⅲ-2

Experiment Ⅲ-2 is an area of interest (AOI) experiment. At this stage, the twelve images that had the highest scenic beauty scores (based on Experiment Ⅰ) for each spatial subtype were selected from 158 samples. To facilitate comparison, the 12 selected images were randomly paired in groups of four, resulting in 12 teams referred to as D₁–D₁₂. A pseudo-randomized order was used, resulting in 12 groups, ensuring that each image had a 1/12 chance of appearing in any corner of the screen to minimize the influence of participants’ viewing habits on the experimental data. Each of the four images in each sample represents an independent AOI distributed in four quadrants.

Subsequently, 37 participants were invited to participate in this experiment. The following four indicators were recorded for the AOI: AOI Total Dwell Time (it may also be known as AOI Total Fixation Duration in certain research), AOI First Fixation Duration, AOI First Fixation Time, and AOI Fixation Count.

2.4. Comparative Analysis

Firstly, using SPSS 27.0 software, data analysis was carried out. Experiment Ⅰ underwent reliability testing, and then Type A, B, and C underwent normalcy testing. Random stratified sampling and the “σ principle” were applied to the classification of scenic beauty qualities.

Secondly, after completing the experiment, the subjective evaluation factors in the SD technique were subjected to correlation analysis. If no significant link was discovered between the factors, multiple linear regression analysis was employed to summarize the scenic beauty perception methods. If there were substantial relationships between the factors, factor analysis was performed for dimensionality reduction, followed by summarization of the scenic beauty perception methods.

Thirdly, one-way ANOVA was applied to analyze the differences between the indicators of different types of TAG, analyzing fixation heatmaps, fixation trajectory maps, and maximum fixation duration maps. The one-way ANOVA, numerical comparison, and graphical comparison techniques were used in the single-image eye-tracking experiment (Experiment Ⅲ-1). It focused on three different TAG spatial types: Type A, Type B, and Type C, and conducted classification and intra-category comparative analyses on 37 samples (37 images). It should be pointed out that 3 of the 40 images used in this stage of the experiment were utilized for preliminary testing and were not included in the statistical analysis. The AOI eye-tracking experiment (Experiment Ⅲ-2) also uses these methods in order to investigate the TAG subtypes perceived to have the highest scenic beauty.

Finally, the link between eye-tracking indicators and subjective evaluation criteria was investigated to determine the factors’ relationship with viewing behaviors. Considering that the public will not typically compare many images during actual visits, in order to make the evaluation method more practical, the six indicators from the previous Experiment Ⅲ-1 were selected for analysis along with the scenic beauty values. As the significance of the impact of these six indicators on the scenic beauty values was unknown, a multiple linear regression analysis was not directly applicable. Instead, this study employed a stepwise regression method, using the six indicators as independent variables and scenic beauty values as the dependent variable. This approach allowed the elimination of weaker influencing factors and the extraction of stronger ones. In addition, the eye-tracking evaluation methods were summarized using a stepwise regression method. A correlation analysis was performed between scenic beauty ratings, subjective evaluation scores, and eye-tracking assessment scores to determine whether the three evaluation techniques were consistent.

3. Results

3.1. Scenic Beauty Perception Based on SBE and SD Methods (Subjective Evaluation)

3.1.1. Scenic Beauty Perception Based on the SBE Method (Experiment Ⅰ)

In Experiment Ⅰ, the subjective assessment of scenic beauty perception was carried out using the SBE method.

Group statistics of scenic beauty value were also performed. The architectural Vertical Greening Sample (Type A) had the lowest average scenic beauty value (SBE_A = 0.2976), the Indoor Greening Sample (Type C) had the highest average scenic beauty value (SBE_C = 0.5889), and the architectural Horizontal Greening Sample (Type B) had the lowest average scenic beauty score (SBE_B = 0.5097). This implies that participants viewed interior greening as having a higher scenic beauty from a macro probability standpoint, whereas architectural vertical greening was thought to have less scenic beauty. The results of normality tests showed that the data had a normal distribution, with significant values for each sample type’s scenic beauty scores being greater than 0.05.

Table 4. Scenic beauty value classification and extracted cases.

TGA Type	Statistical Interval	Scenic Beauty Value			Number of Cases	Normal Distribution, Boxplot, and Scatter Plot
		Low	Medium	High
Vertical Greening Systems	Value Range	<−0.1560	−0.1560~0.7512	>0.7512	-
	Total Quantity	9	24	5	38
	Extracted Quantity	1	7	2	10
Horizontal Greening Systems	Value Range	<0.2301	0.2301~0.7893	>0.7893	-
	Total Quantity	24	60	14	81
	Extracted Quantity	3	12	5	20
Indoor Greening	Value Range	<0.3749	0.3749~0.8029	>0.8029	-
	Total Quantity	6	25	8	39
	Extracted Quantity	2	7	1	10

shows the number of cases using the “σ principle” of classification and random stratified sampling.

3.1.2. Scenic Beauty Perception Based on the SD Method (Experiment Ⅱ)

Experiment Ⅱ conducted a one-way ANOVA on the TAG kinds and their accompanying indicator scores, and used the SD method to provide a subjective assessment of scenic beauty perception. The functional integration factor was eliminated from the one-way ANOVA as it did not pass the homogeneity of variance test. Following this exclusion, all of the other indicators’ significance values were higher than 0.05, suggesting that TAG type had no discernible impact on the subjective rating components or scenic beauty values. This implies that no single factor dominates the impression of scenic beauty, even while the TAG type affects each factor. In other words, regardless of the TAG type, as long as the design is reasonable, its scenic beauty can be well perceived by the public.

3.1.3. Scenic Beauty Perception Evaluation Based on Subjective Experiment

A subjective assessment formula for the perception of scenic beauty perception was developed by combining the findings of Experiments Ⅰ and Ⅱ. The subjective evaluation factors were subjected to correlation analysis, which demonstrated that direct linear regression analysis was not practical and that the correlations between the factors were significant. In order to reduce the seven subjective evaluation factors to a lower number of variables, factor analysis was utilized. According to the computation, the dataset was appropriate for factor analysis as the KMO value was 0.871, which is higher than 0.7, and the Bartlett’s test of sphericity revealed a significance value of p < 0.001. To identify common factors, principal component analysis (PCA) was employed. As shown in

Table 5. Total variance explanation (principal component analysis).

Common Factor	Initial Eigenvalues			Extracted Sums of Squared Loadings			Rotated Sums of Squared Loadings
	Total	Percentage of Variance	Cumulative Percentage (%)	Total	Percentage of Variance	Cumulative Percentage (%)	Total	Percentage of Variance	Cumulative Percentage (%)
1	4.769	68.131	68.131	4.769	68.131	68.131	4.379	62.558	62.558
2	1.604	22.919	91.050	1.604	22.919	91.050	1.994	28.492	91.050
3	0.226	3.223	94.273
4	0.163	2.327	96.600
5	0.095	1.355	97.955
6	0.079	1.135	99.089
7	0.064	0.911	100.000

, there were two common factors with initial eigenvalues greater than 1. Common factor 3, along with subsequent factors, had initial eigenvalues less than 1, indicating weak explanatory power for the original indicators, and was, therefore, excluded. Thus, two common factors were isolated, accounting for a cumulative value of 91.050%. These factors can be used in place of the original seven variables in the scoring of the photographs, and they offer a good explanation of the overall subjective evaluation factors.

As shown in

Table 6. Component score coefficient matrix.

Common Factor	Landscape Subjectivity	Scene Balance	Vegetation Layering	Landscape Coordination	Architectural Coordination	Functional Integration	Color Coordination
F1	−0.112	0.235	−0.067	0.192	0.247	0.217	0.209
F2	0.542	−0.082	0.502	0.062	−0.120	−0.038	0.019

, the two common factors in the rotated component matrix were labeled as F₁ and F₂, referred to as the subjective evaluation dimensions. F₁ was given the name “Spatial Coordination” because of its high loadings on scene balance, landscape coordination, architectural coordination, functional integration, and color coordination. F₂ was dubbed “Ecological Stratification” because it demonstrated considerable loadings on vegetation layering and landscape subjectivity. To make it easier to summarize the calculation formulas, the subjective assessment factors were marked as X₁ to X₇. The common factors F₁ and F₂ are calculated using the following formulas (Formula (2) and (3)):

F₁ = −0.112 X₁ + 0.235 X₂ − 0.067 X₃ + 0.192 X₄ + 0.247 X₅ + 0.217 X₆ + 0.209 X₇

(2)

F₂ = 0.542 X₁ − 0.082 X₂ + 0.502 X₃ + 0.062 X₄ − 0.120 X₅ − 0.038 X₆ + 0.019 X₇

(3)

Finally, the subjective evaluation formula for scenic beauty perception was derived based on factor analysis (Formula (4)). The variance contribution rates corresponding to F₁ and F₂ were used as weights. The SBE value for sample i using the subjective evaluation method is represented by SBE_Si in this formula; the indicator weights are 62.558 and 28.492, the total weight value is 91.050, and the scores for ecological stratification and spatial coordination are, respectively, represented by F₁ and F₂.

SBE_Si = (62.558 F₁ + 28.492 F₂)/91.050

(4)

3.2. Scenic Beauty Perception Based on Eye-Tracking Experiment (Objective Evaluation)

3.2.1. Single-Image Eye-Tracking Experiment (Experiment Ⅲ-1)

(1)

Classification Comparative Analysis

The classification comparative analysis employed one-way ANOVA to analyze the eye-tracking data of the 37 samples (37 images) across different spatial types and investigate the objective influence of spatial type on scenic beauty perception. The findings revealed that the three varieties of TAG differed significantly in two indicators: Total Fixation Duration and Saccade Count.

• Total Fixation Duration: Types A and B had significantly shorter Fixation Duration than Type C, indicating that Indoor Greening’s integration with building functionality helped participants relate to daily life scenarios, resulting in stronger attention paid during observation. On the other hand, there were fewer cases in horizontal and vertical surface greening where greening was integrated with building functionality, and participants had less life experience in this regard.
• Average Fixation Duration: Average Fixation Duration for Type C was significantly higher than for the other types, indicating that participants paid more attention to the design details of Indoor Greening and had a stronger interest in it. In the Vertical Greening System and the Horizontal Greening System, the participants primarily understood the design through an overall scene perception, with less attention paid to specific design details during limited observation time.
• First Fixation Duration: First Fixation Duration for Type B and Type C showed significant differences, suggesting that the public required the longest initial information processing time for the Horizontal Greening System, possibly because this type of landscape was relatively complex. A longer First Fixation Duration generally indicates that the image had a stronger appeal.
• Fixation Count: There is no significant difference between Types A, B, and C in Fixation Count (p > 0.05).
• Saccade Count: Type C had the lowest Saccade Count, indicating that the information complexity in Indoor Greening images was lower and easier to understand. In contrast, images of vertical and horizontal greening, with a greater amount of information, required more saccades for participants to gather the necessary data.
• Average Saccade Amplitude: There is no significant difference between Types A, B, and C in Average Saccade Amplitude (p > 0.05).

(2)

Within-Group Comparison Analysis

The within-group comparison analysis used numerical and graphical comparison methods to examine the eye movement data from 37 samples (37 images) of the same spatial type. It aimed to determine the objective influence of design elements on scenic beauty perception. In the Vertical Greening System, Sample A₁₂ had the longest Total Fixation Duration; A₁₃ had the longest First Fixation Duration and the highest Average Saccade Amplitude; and A₁₄ had the shortest Average Fixation Duration, the largest Fixation Count, and the largest Saccade Count (Figure 5,

Table 7. Fixation heatmap of representative Vertical Greening System samples.

Representative Vertical Greening System Samples Fixation Heatmap
A12	A13	A14
Legend

). More specifically, in Sample A₁₂, the architectural volumes on both sides of the pedestrian bridge were visually enhanced with vertical greening, shattering the sense of large-scale building mass and providing interest and layering, consequently optimizing the public viewing experience. In Sample A₁₃, the building was covered with extensive greening, and while the initial visual effect was high, the haphazard plant arrangement resulted in poor overall aesthetic quality, causing participants to lose interest in viewing. Sample A₁₄ demonstrated a basic and easy-to-understand vertical greening design for the external corridor; but, due to the lack of variety in the greening, participants lost interest after first grasping the image’s content, with fixations becoming more dispersed.

In the Horizontal Greening System, B₇₃ had the longest First Fixation Duration, whereas B₁₄ was second in Total Fixation Duration, had the longest Average Fixation Duration, the smallest Fixation Count, and the smallest Saccade Count (Figure 6,

Table 8. Fixation heatmap of representative Horizontal Greening System samples.

Representative Horizontal Greening System Samples Fixation Heatmap
B73	B14	B78
Legend

). Specifically, in B₇₃, the roof greening with an undulating curve design drew participants’ heightened attention. During the trial, participants were drawn to the undulating roof form and thought the design was unusual, visually appealing, and had strong overall coherence. In B₁₄, participants mainly focused on the roof stairs and shrubbery design, particularly the upper section of the stairs where a child appeared in the image. Due to limited information and a relatively simple design, participants gradually lost interest in further exploration.

In Indoor Greening, C₃₇ exhibited the largest saccades and the smallest Average Saccade Amplitude. Participants were drawn to C₃₇’s intricate and multi-layered greening scheme. C₂ had the highest Total Fixation Duration, the lowest First Fixation Duration, the longest Average Fixation Duration, the smallest Fixation Count, and the smallest Saccade Count (Figure 7,

Table 9. Fixation heatmap of representative Indoor Greening samples.

Representative Indoor Greening Samples Fixation Heatmap
C37	C2	C11
Legend

). This illustrates how, during the spatial experience process, the public first took in the overall picture, noting the interior design’s aesthetic appeal and color scheme, and then they took in the greening, especially growing a sense of intimacy and desire for the seating arrangement that was in close proximity to the greenery. Eye movement data also revealed that participants made multiple and detailed observations. In C₂, the greening design in the office atrium was densely layered, creating a visual focal point. Participants noted beauty in the lighting, color harmonizing, and composition. Subsequently, they observed that the arrangement of plants in the atrium was reasonable and harmonized with the office atmosphere.

3.2.2. AOI Eye-Tracking Experiment (Experiment Ⅲ-2)

The AOI Eye Movement Experiment (Experiment Ⅲ-2) employed one-way ANOVA (

Table 10. Variance analysis table of eye-tracking data for AOI.

Dependent Variable	Sum of Squares	Mean Square	F	p
AOI Total Dwell Time	6.865	0.624	9.095	<0.001
AOI First Fixation Duration	0.033	0.003	1.097	0.391
AOI Fixation Count	49.438	4.494	3.971	<0.001
AOI First Fixation Time Score	1562.500	142.045	0.142	0.999

), numerical comparison, and graphical comparison methods (

Table 11. Eye-tracking experiment chart of AOI for representative samples.

Phase	Analysis	D6	D12	…
TAG AOI Experiment	Fixation Heatmap			…
	Legend
	Fixation Trajectory Map			…
	Legend
	Maximum Fixation Duration Map			…
	Legend

, Figure 8) to analyze the eye movement data of 12 composite samples (12 sets of square grid images). It aimed to screen the TAG subtype with the highest scenic beauty. Table 11 presents the representative samples D₆ and D₁₂. In D₆, the top left picture represents Platform Greening, the top right picture represents Balcony Greening, the bottom left picture represents Indoor Greening, and the bottom right picture represents Climbing Greening. In D₁₂, the top left picture represents Indoor Greenhouse Greening, the top right picture represents Artificial Substrate Greening, the bottom left picture represents Terraced Greening, and the bottom right picture represents Plateau Greening. Through the use of warm and cool colors, circle colors, circle sizes, and number sequences, the fixation intensity, Fixation Durations, and annotation trajectories of different participants were labeled, superimposing the fixation heatmap, fixation trajectories, and maximum Fixation Duration.

The experimental results of Experiment Ⅲ-2 are presented in the form of visual images. This study organized the data for each indicator and performed a one-way ANOVA. It was found that there were significant differences in the Total Fixation Duration and the number of fixations in the AOI among the 12 different TAG subtypes (p < 0.05).

• AOI Total Dwell Time: The AOI Total Dwell Time is determined by an individual’s level of interest, attention to the area of interest’s content, and communicative context-specific demands. In situations requiring intense focus or interest, an individual’s overall fixation time in the area of interest increases. After analyzing the AOI Total Dwell Time data and conducting post-experiment interviews, it was discovered that unique design forms quickly attract participants, but traditional greening approaches frequently fail to capture their interest. In the experiment, the inclined and terraced greening are uncommon in daily life, and their designs are rather unusual for participants, with a variety of design styles, making them more likely to pique people’s curiosity. Artificial Substrate Greening, on the contrary, is quite simple in design, often consisting of a ring of planting that surrounds the building. This repetitive design makes it harder to attract people’s attention.
• AOI Fixation Count: Based on the AOI Fixation Count data combined with post-experiment interviews, it was found that the participants showed a higher level of interest in Terraced Greening, Platform Greening, and Inclined Greening. This is because these examples show a greater connection of greening and architectural functionality. The public tends to focus more on images that depict specific functional purposes, as they can link to real-life experiences and better understand the space.

3.2.3. Scenic Beauty Perception Evaluation Based on Eye-Tracking Experiment

The stepwise regression findings showed that the data satisfied the independence criteria (R² = 0.572, p < 0.001, Durbin–Watson = 1.716). Strong relevance was indicated by the Fixation Count (FC_i) and Saccade Count (SC_i), both of which displayed significance with p < 0.001 (

Table 12. Stepwise regression model coefficients (dependent variable: scenic beauty value).

Variable Name	Unstandardized Coefficients		Standardized Coefficients
	B	Standard Error	Beta	t	p
Constant	−0.683	0.677		−1.009	0.320
Fixation Count	0.227	0.037	0.744	6.145	<0.001
Saccade Count	−0.174	0.036	−0.588	−4.858	<0.001

). Consequently, the formula for determining scenic attractiveness (Formula (5)) was developed as follows:

SBE_Oi = −0.683 + 0.227 × FC_i − 0.174 × SC_i

(5)

SBE_Oi represents the scenic beauty value for sample i, while the eye-tracking evaluation indicators of sample i are represented by FC_i and SC_i. The indicators have weights of 0.227 and −0.174, respectively, while the constant value is −0.683 (Table 12).

3.3. Comparative Analysis

Experiments Ⅰ and Ⅱ involved scoring single images, which is consistent with Experiment Ⅲ-1. Therefore, for the comprehensive analysis, the data from the six indicators of Experiment Ⅲ-1 were selected and correlated with the data from Experiments Ⅰ and Ⅱ. This analysis aimed to examine the feedback from the experimental data from the perspective of landscape elements, in order to derive design strategies.

3.3.1. Mutual Validation of Subjective and Objective Evaluation Systems

The subjective assessment formula (formula 4) and the eye-tracking evaluation formula (formula 5) were used to independently generate the SBE_Si and SBE_Oi values for each of the 37 distinct images. These values were then correlated with SBE_i. As shown in Figure 9, the findings demonstrated that all pairs had a substantial positive correlation, with the significance between the SBE_i, SBE_Si, and SBE_Oi values being less than 0.001. The three values had a high degree of agreement, suggesting that the findings could support one another. Both assessment techniques showed excellent accuracy and were mostly in line with the findings of the public’s direct evaluation.

It should be noted that the predetermined scoring range of the Likert 5-point scale (−2, −1, 0, 1, 2) used in Experiments Ⅰ and Ⅱ is the source of the negative values seen in this investigation. The mathematical characteristics of the subjective evaluation formula for scenic beauty perception based on factor analysis (Formula (4)) may cause localized deviations in computational results for particular samples, which explains the sporadic extreme negative values seen in A₁₃ and B₆₅. The presence of such outliers aligns with theoretical expectations, and correlation analyses confirm that the scores from the three methodologies exhibit a positive correlation across the overall distribution (p < 0.001).

3.3.2. Correlation Analysis Between Subjective Evaluation Indicators and Eye Movement Indicators

A correlation analysis was conducted between the subjective evaluation dimensions and the data from the six indicators, and the results were plotted as a heatmap (

Table 13. Heatmap of correlation analysis between subjective factors and six eye-tracking indicators.

Indicator Name	Total Fixation Duration	Average Fixation Duration	Fixation Count	First Fixation Duration	Average Saccade Amplitude	Saccade Count
Ecological Stratification Score
Spatial Harmony Score
Legend

* Indicates a significant correlation at the 0.05 level (two-tailed). ** Indicates a significant correlation at the 0.01 level (two-tailed).

). Upon observing the heatmap, the following conclusions were drawn:

• Four eye movement indicators (Total Fixation Duration, Fixation Count, First Fixation Duration, and Average Saccade Amplitude) had a significant correlation with Spatial Coordination. However, Ecological Stratification had a significant correlation with only one eye movement indicator (Average Saccade Amplitude).
• Both Ecological Stratification and Spatial Coordination showed a significant negative correlation with Average Saccade Amplitude. However, the significance of the correlation between Ecological Stratification and Average Saccade Amplitude was stronger than that between Spatial Coordination and Average Saccade Amplitude.
• Total Fixation Duration, Fixation Count, and First Fixation Duration all showed a significant positive correlation with Spatial Coordination.
• A further analysis of the highly correlated indicators at the 0.01 significance level reveals that Total Fixation Duration and Fixation Count both showed a significant positive correlation with Spatial Coordination, while Average Saccade Amplitude only showed a significant negative correlation with Ecological Stratification.

4. Discussion

4.1. Systematic Scenic Beauty Perception Evaluation of TAG

TAG involves the integration of architectural spaces and greening landscapes, and the complexity of its scenic beauty perception evaluation makes the research challenging. Previous landscape visual quality studies have mostly focused on the investigation and analysis of several landscape nodes within one to three project cases, with few studies systematically evaluating a specific type of landscape. This study used standard images as the medium and various kinds of approaches to acquire TAG samples to create a 300-case library that covered a wide range of TAG projects. Through screening and classification, 158 samples (158 images) (the ratio and precision of each image were harmonized using Photoshop to fully capture the theme effect of TAG) were selected for Experiment Ⅰ, followed by further experimentation with 40 samples (40 single images) for Experiment Ⅱ, and 52 samples (40 single images and 12 groups of AOI images) for Experiment Ⅲ. The comparison results showed that the subjective and objective assessment outcomes were highly comparable, indicating that both methods possess certain scientific value. A series of specific research can be carried out in the future as the database expands, such as a systematic evaluation of beauty perception for different forms of building three-dimensional greening.

4.2. Subjective and Objective Evaluation Methods

This study includes two sorts of systematic evaluations for the visual perception of TAG scenic beauty: subjective and objective. The subjective evaluation includes SBE and SD approaches, whereas the objective evaluation relies on eye-tracking experiments. These three approaches, which are based on standardized images, are quick and easy to use, compensating for the drawbacks of conventional techniques like indirectness, ineffectiveness, and time-consuming information gathering. They also provide a strong basis for the real-world implementation of classification prediction. In addition to being applicable to an extensive range of application scenarios, these evaluation paths—subjective evaluation and objective evaluation—can also be combined to enhance one another and offer scientific references for future TAG designs and spatial decision-making.

Based on standardized image data collection and quantitative analysis, the findings of visual perception measurement of TAG scenic beauty perception supported by several technical approaches (SBE, SD, eye-tracking experiment method) were compared and analyzed. Correlation analysis of the subjective evaluation scores, eye-tracking evaluation scores, and scenic beauty values shows significant positive correlations between each pair of evaluation methods. This indicates that the subjective evaluation results, direct public evaluation, and the objective physiological data analysis results are generally consistent, confirming the subjective evaluation method’s validity from multiple perspectives. The SBE technique focuses on efficiency and standardization, whereas the SD method emphasizes multidimensional analysis. Eye-tracking investigations can provide objective behavioral data that reveal visual preferences. In practical applications, single or mixed approaches might be chosen based on research goals and resource availability. The scenario adaptability of the three methods is as follows:

The SBE approach is ideal for swiftly evaluating a large number of landscapes and extracting essential design characteristics. When undertaking an in-depth examination of subjective perception aspects or developing thorough evaluation models, the SD approach should be prioritized. Eye-tracking experiments work well for examining visual behavior mechanisms or detailed feedback in laboratory settings.

Previous research on landscapes visual evaluation relied solely on subjective evaluation methods or objective evaluations based on eye movement experiments, with less synchronous use of subjective and objective approaches or a combination of both. The evaluation approach offered in this study can be used to assess the scenic beauty of TAG based on the applicable scenarios, making it easy to obtain the public’s perspective of the beauty of specific designs. These approaches can be combined according to their appropriate scenarios, as well as employed independently. The evaluation system can shift from a “spatial efficiency priority” to a “human-oriented perception driven” approach, offering a scientific foundation for creating a people-centered urban landscape.

4.3. The Interaction Between Landscape Elements and Scenic Beauty Perception of TAG

In order to further explore the correlation between public’s visual behavior and scenic beauty perception of TAG, this study conducted a correlation analysis between the data of subjective evaluation dimensions and six indicators. The interactive relationship of “Landscape Elements—Scenic Beauty Perception of TAG” has also been confirmed by objective evaluation (eye-tracking experiment):

First, Spatial Coordination is far more crucial than Ecological Stratification among the subjective rating indicators. Second, Ecological Stratification is a more significant metric for affecting how the general public perceives the architectural vertical greening’s aesthetic elements. A multi-level biological landscape will draw public attention to its details; therefore, in eye-tracking experiments, subjects’ eyes will not rapidly broaden the search region, but will focus on the details. Finally, Spatial Coordination is a major factor influencing the public’s understanding and cognition of landscapes. Images with harmonious compositions, greater coordination between buildings and their surrounding environments, and better integration of building functions and greening tend to have richer design information, leading to a higher perception of aesthetic value by the public.

4.4. Limitations of the Study

This study acquired samples using field surveys, book extracts, and data retrieved from websites, and picture source constraints influenced the database’s breadth. With the increasing prevalence of TAG, Internet platform scraping technology may be used to increase the sample size of the case database, providing more abundant samples for scenic beauty research and improving the scientific nature of the study. Furthermore, the process of perceiving beauty is intricate and influenced by a number of outside variables, such as individual subjective variations, contextual elements, etc. Knowledge, experience, interests, and hobbies may all influence how subjects perceive beautiful scenery. Physical environmental elements like illumination, brightness, and the size of the observation area can all influence how individuals perceive appealing scenery overall. Because of this, the study’s experimental stimuli were static images, which made it impossible to account for the influence of dynamic surrounding components. Future studies could include collecting panoramic images and conducting VR-based eye-tracking experiments, incorporating the influence of environmental soundscapes to broaden the participants’ visual perception and physiological experiences, and developing a series of targeted expansion studies.

5. Conclusions

• To explore the perception of the beauty of three-dimensional greening in buildings, this study gathered 300 cases and created a case library. Typical samples were chosen at random to participate in the study, ensuring that all types of cases were represented in the experiments. Three experiments were set up in stages to assess perceptions of scenic beauty. Multiple methodologies, including questionnaire surveys, factor analysis, and data analysis, were employed in systematic research, and a systematic evaluation model was developed. This study suggests two evaluation paths—subjective and objective—to gauge how the general public perceives TAG using factor analysis and stepwise regression. This study will verify the subjective evaluation system with the objective evaluation system to ensure the evaluation’s efficacy and rationality. Additionally, correlation analysis between subjective evaluation indicators and eye-tracking indicators will be conducted to investigate the primary factors influencing the public’s perception of the beauty of TAG, providing scientific reference for the construction of urban landscapes.
• With the technological revolution of lightweight devices and open source algorithms, physiological data collection will increasingly be used to evaluate urban spaces. Future research may focus on the collaborative mechanism between landscape aesthetic value and ecological function, and promote the transformation of the evaluation system from “spatial efficiency priority” to “human-oriented perception driven”. The study conducted a “landscape-element” analysis based on subjective and objective experimental data and built a linkage of “evaluation experiment–data feedback–design guidance”. The evaluation approaches proposed in this study demonstrate the benefits of high efficiency and convenience in evaluating the creation of landscape atmosphere through images, leading the way for a new path for perception, evaluation, and optimization of TAG designs.
• Through the comparison of subjective and objective assessments, this study determined the interaction between spatial elements and public perception of beauty, offering a scientific basis for the three-dimensional greening of urban systems: In TAG design, the importance of Spatial Coordination is significantly greater than that of Ecological Stratification. Focusing on Spatial Coordination design can enhance the overall aesthetic appeal to a greater extent, while emphasizing Ecological Stratification can invigorate the details of the design to a greater degree.