This study employs the PRISMA method for conducting a systematic scoping review, aiming to enhance the transparency and reproducibility of the literature screening process. The literature searches spans five databases: Web of Science, ScienceDirect, Scopus, ACM Digital Library, and A&HCI. The PRISMA approach ensures the structural rigor of the research process while increasing the transparency and consistency of the review process. This systematic scoping review adheres to the guidelines outlined in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR).

The literature search was conducted on November 20, 2024, using search terms including “image,” “image vision,” “computer vision,” “cultural,” “cultural transmission,” “cultural communication,” “public,” and “service.” The search strategy included the following steps: (1) a preliminary screening in all databases to exclude irrelevant studies; (2) removal of duplicate references using EndNote 21.4 software; (3) detailed screening of titles and abstracts to identify studies relevant to the research topic; and (4) full-text review to exclude studies that were not directly related to the research topic. The specific search formula is shown in Table 1 (The PRISMA-ScR checklist is provided in Appendix B).

All retrieved records were imported into the reference management software EndNote version 21.5, and duplicate entries were removed using the software. Two independent reviewers (JY and TL) screened the titles and abstracts of the articles based on predefined inclusion criteria. In cases of discrepancies between the reviewers, a third reviewer (PP) was consulted to reach a consensus. The inclusion criteria (see Table 2) for selecting studies were as follows: (1) research involving the application of VI in CT; (2) involves the specific application of CT and inheritance; (3) VI studies based on AI technology; (4) studies presenting original research; (5) published between 2015 and 2024; (6) full text in English.

A data extraction form was developed based on the scoping review methodology guidelines provided by the Joanna Briggs Institute. After conducting a preliminary trial with five articles, the form was revised to enhance its effectiveness and accuracy. The form includes the following key data extraction items: author, publication year, country, AI technology, topic, VI type, target group, dissemination platform, dissemination path, and findings. Data extraction was performed by two independent reviewers, and any discrepancies were resolved through consultation with a senior reviewer to ensure the accuracy and consistency of the extracted data.

During this process, all three authors concurred on narrowing the scope of categories to ensure that they specifically address the research questions of this study. Descriptive statistics were employed to systematically organize, summarize, and report the results. The findings were presented in a narrative format, complemented by figures and tables to enhance clarity.

Figure 2 illustrates the annual evolution trend of the included literature. The time series indicates that relevant research began in 2015, gradually developing thereafter, and reaching a peak in 2022 with a total of four studies published. During the COVID-19 pandemic, the closure of cultural institutions and the global shift to online spaces prompted a transition of cultural content toward virtual engagement. This increased the need to explore the role of digital and virtual media in CT, which may explain the rise in publications in 2022 and the following year (Li et al., 2022).

Regarding the geographical distribution of the research, Figure 3 displays the number of contributions by country and region. China published six papers, accounting for 33.3% of the total, temporarily leading the field. China has 56 cultural and natural heritage sites listed on the UNESCO World Heritage List, ranking second globally. This leadership is likely closely related to the rapid development of China’s digital cultural industry in recent years and government support for visual culture research. At the regional level, Asia dominates with ten articles, representing 55.6%, followed by Europe with six articles (33.3%), and North America with two articles (11.1%).

In the included studies, the majority employed DL algorithms for AI-based VI processing in CT applications, particularly CNN. 11 studies utilized CNN to process and generate image data. CNN has significant advantages in handling image data, enabling tasks such as image classification and detection. Secondly, cloud computing was used for data processing in two studies. Image processing often involves textual data; for example, input text data can be transformed into VI through generative AI, thereby involving LLM and NLP. However, HCI and XAI were less frequently employed in the research (see Figure 4).

The thematic classification identified 18 topics across seven major categories. CES evaluation and utilization was the most prominent, encompassing five topics, highlighting researchers’ focus on assessing and applying CES in various contexts, essential for urban planning and environmental management. Landscape & environmental management, as well as urban image & perception, included four and two topics, respectively, emphasizing the role of landscape management and urban image in enhancing sustainability and residents’ quality of life. Classification systems and recreation & aesthetic evaluation each comprised two topics, indicating the need for specific classification methodologies and aesthetic value assessments. Cultural heritage and technology in urban Studies featured one and two topics, respectively. This reflects the increasing recognition of cultural heritage tourism and emerging technologies, such as hybrid 3D virtual museums and generative AI, in enriching cultural experiences and research. Overall, the distribution of topics underscores current academic interests in CES, landscape management, and technological applications (see Table 4).

The analysis reveals that social media users (36%), tourists (28%), and urban residents (22%) are the primary target groups across the studies. The frequent combination of these groups highlights the importance of social media in urban and tourism research, as well as the need to understand the dynamics between local residents and visitors. Additionally, the inclusion of niche groups points to diverse research interests that can be further explored to enrich the understanding of cultural and urban ecosystems (see Figure 5).

The analysis reveals that Flickr and social media are the predominant dissemination platforms, with usage frequencies of 41 and 35%, respectively. This emphasizes the importance of visual content and social interactions in information dissemination. The strategic combination of multiple platforms underscores the need to reach a broader and more varied audience. While emerging technologies are currently underutilized, their potential for enhancing dissemination strategies is evident, presenting opportunities for future research and application. The integration of physical and digital dissemination methods also highlights the evolving landscape of information sharing, aiming for more comprehensive and effective communication (see Table 5).

The analysis reveals that digital dissemination (50%) is the predominant pathway for CT, highlighting its central role in contemporary cultural exchange. The strategic combination of digital dissemination with localization adaptation (22%) and social media dissemination (11%) underscores the importance of contextual and interactive approaches in enhancing the effectiveness of CT. Cross-cultural dissemination (11%) also plays a notable role, reflecting the interconnectedness of global cultures. Cultural inheritance (6%) maintains a unique position, emphasizing the preservation of traditional cultural elements (see Figure 6).

The analysis of the findings reveals several key themes underpinning the mechanisms of CT within urban contexts. Social media images emerge as a predominant factor, influencing both urban image and cultural cognition across various domains such as urban parks, building functions, and coastal landscapes. Specifically, SMP facilitates the dissemination and enhancement of cultural understanding through visual content, as evidenced by multiple findings.

Image classification and recognition technologies play a crucial role in supporting CES evaluation and promoting broader cultural dissemination. These technologies enhance the efficiency and accuracy of disseminating cultural landscapes and intangible cultural heritage, leveraging tools such as computer vision, open-source classifiers, and multi-modal image classification. DL further augments CT by generating immersive narratives and facilitating aesthetic dissemination. The integration of advanced technologies like LiDAR, remote sensing, and 3D models enhances the depth and reach of cultural dissemination efforts, enabling more comprehensive and interactive cultural experiences. Cultural flagship events significantly contribute to the reconstruction of urban images and the enhancement of cultural identity. The use of unlabeled images presents both advantages and limitations, indicating a need for balanced approaches in CT strategies. Overall, the findings underscore the synergistic interplay between social media, advanced image technologies, and DL in fostering effective CT.

DL enhances CT by generating immersive narratives and promoting aesthetic dissemination. Studies show that AI-generated narratives and images that visually present users in a story context can achieve good immersion and engagement (Dong, 2025). Image classification and recognition technologies play a crucial role in supporting CES evaluations and promoting broader cultural dissemination (Ju, 2024). DL improves the efficiency and accuracy of disseminating cultural landscapes and intangible cultural heritage by leveraging tools such as computer vision and multimodal image classification (Gîrbacia, 2024). Pre-trained CNNs have shown high effectiveness in accurately identifying cultural heritage elements from social media images (Belhi et al., 2021). For instance, some studies have used pre-trained VGG19 and Xception models to significantly improve the accuracy and efficiency of traditional cultural heritage image classification through transfer learning (Janković Babić, 2024). The integration of advanced technologies such as LiDAR, remote sensing, and 3D modeling enhances the depth and breadth of CT by creating interactive cultural experiences for urban populations (Y. Li et al., 2023), as such technologies enable computers to capture in-depth environmental data and generate artistic output based on sensing. A hybrid 3D virtual museum combines panoramic images and models to offer a more realistic and interactive cultural experience (Barrile et al., 2022).

Social media images, through a collective construction process, significantly influence cultural cognition and the formation of urban identity across different urban contexts (Loughran et al., 2015). Unlike the unidirectional dissemination of traditional media, SMP facilitate two-way cultural exchange through user-generated content and interactive functions (such as comments, reposts, and likes), encouraging users to actively participate in content creation and sharing (Eroglu, 2023). This collective construction process significantly influences cultural cognition and enhances the role of visual content in promoting cultural understanding (Gooding, 2004). Moreover, real-time content dissemination on SMP significantly accelerates CT relative to the slower, fixed schedules of traditional broadcast media (Chukwu, 2023). On a technical level, image classification and recognition technologies play a crucial role in supporting CES evaluations and promoting broader cultural dissemination. These technologies, utilizing tools like computer vision and multimodal image classification, improve the efficiency and accuracy of transmitting cultural landscapes and intangible cultural heritage. Research indicates that platforms like Flickr and Weibo are core channels for disseminating visual content and facilitating social interaction (Liang et al., 2022).

VI, due to its intuitive and universal nature, can transcend language barriers, simplifying complex cultural concepts and promoting understanding without the need for translation (Vishwakarma, 2023). For example, traditional Chinese New Year pictures express themes of happiness, good fortune, and prosperity through symbols like figures, animals, and plants, which helps audiences from different cultural backgrounds understand core Chinese cultural values (Welch, 2013). Algorithmic diversification strategies, such as “explore mode” and randomized recommendations (Fang et al., 2020), are effective in broadening users’ exposure to cross-cultural content by correcting the “filter bubble” effect often found in recommendation systems (Grossetti et al., 2021).

Furthermore, AI-driven visual technologies are pivotal in creating immersive and interactive cultural experiences that drive cross-cultural communication. In the context of digital cultural tourism, technologies like virtual tours and 3D modeling allow global audiences to explore cultural heritage sites and traditions regardless of physical distance (Napolitano et al., 2018). These visual mediums not only attract tourists but also serve as educational tools, enabling a deeper appreciation of diverse cultures. The application of AI in this domain transforms passive viewing into an active, engaging experience, fostering a more direct and personal connection between individuals and foreign cultures.

Despite advancements, AI technology introduces several challenges in CT. A primary concern is algorithmic bias, as many AI models are trained on datasets dominated by Western aesthetic standards, leading to biases in recognizing non-Western cultural content. This can distort cultural heritage representations and exacerbate cultural inequalities (Gosal and Ziv, 2020). AI models often lack a deep understanding of cultural symbols and contexts, resulting in misinterpretations, especially with religious or historical images (Fan et al., 2023). Another significant risk is cultural homogenization and the loss of creativity. In creative fields, reliance on repetitive data can stifle innovation, leading to homogenized cultural products (Fan et al., 2023).

Privacy and data security also present pressing concerns. Training facial recognition models requires vast amounts of personal images, raising issues of privacy infringement (Su et al., 2023). AI can be misused to create fake images or videos, which can mislead the public and undermine social trust, such as fabricated political videos used as propaganda (SHI et al., 2021). To mitigate these issues, both technological and methodological improvements are essential. Expanding AI training datasets to include diverse cultural elements such as language, music, and text can help reduce bias (Fan et al., 2023). Incorporating cultural context into algorithm design through cultural tagging or building knowledge bases can enhance the understanding of cultural nuances and reduce misinterpretations (Huai et al., 2022). Developing explainable AI algorithms is crucial for improving transparency in decision-making and identifying potential biases (Huai et al., 2022).

Future research should expand its methodological scope to provide more balanced insights into the application of AI and SMP in diverse cultural contexts. This can be achieved by incorporating a wider range of sources from various cultural backgrounds and conducting more nuanced analyses of the practical challenges and ethical implications. Additionally, the potential of digital tools such as VR, 3D modeling, and panoramic imaging to simultaneously strengthen cultural engagement and ecological conservation should be explored. This could be achieved by creating virtual tours of heritage sites that highlight both their cultural significance and the need for ecological protection. Ultimately, future studies should investigate the potential of user-generated content in supporting CES by developing collaborative cultural-ecological conservation models that foster public participation and promote conservation efforts.

This scoping review has several limitations and strengths that warrant discussion. Many studies relied heavily on user-generated content from SMP, which inherently introduces biases. Factors such as sample selection, tagging practices, and subjective interpretations may distort the representativeness of cultural narratives. These limitations are compounded by the under-representation of diverse cultural contexts, particularly from regions with limited access to advanced technological tools. Addressing this imbalance would require broader geographic and demographic inclusion in future research. In conclusion, while this systematic scoping review provides valuable insights into the role of VI in CT, particularly in the era of DL, it is crucial for future research to address these limitations by expanding the scope of the literature review, incorporating a broader range of sources, and providing a more nuanced analysis of the challenges and practicalities of implementing AI and SMP strategies in diverse cultural contexts.