Traditional positioning and navigation technologies have received extensive application, yet the emergence of image-based positioning represents a novel and promising avenue within the field. The incorporation of deep learning has significantly advanced image processing capabilities, thereby enhancing the potential for image-guided positioning applications. This study presents an innovative approach to geographical location prediction directly from image content, employing deep learning models to extract location data. We pre-train our models using open-source datasets, ensuring a robust foundation for subsequent analysis. Our methodology encompasses the use of mobile measurement devices to evaluate natural targets across indoor and outdoor settings, capturing images that, through computational vision, inform target presence and user positioning. The harnessed image geolocation information enables a comprehensive examination of community structures and user behavior within social networks, optimizing service delivery. Furthermore, this technology empowers social platforms to proactively identify and filter content with sensitive geolocation data, safeguarding user privacy and preventing the spread of inappropriate content. The research significantly contributes to the domain of internet security. Experimental outcomes indicate high precision with minimal error in urban street positioning. However, a notable decline in accuracy is observed in outdoor environments, particularly within local scenes. This variance underscores the necessity for ongoing research to bolster the robustness of visual positioning systems in diverse geographical settings.