Amid the surging digital wave, large AI models are reshaping the global technology landscape with disruptive force. Natural language models enable human-machine dialogues with near-real authenticity, vision models unlock the ultimate code for image interpretation, and multimodal models strive to construct "digital twins" in three-dimensional worlds.
Yet, how can these technologies truly take root in vertical industries? As a core component of the digital economy, the geospatial information sector is rapidly integrating cutting-edge technologies like AI large models, continuously empowering intelligent industry applications and advancing toward new heights of geospatial intelligence.
With the meteoric rise of large models such as DeepSeek and ChatGPT, geospatial intelligence software technologies are undergoing explosive iterations. The exploration of GIS agents has entered a critical phase—what disruptive technical architectures will emerge? What product ecosystems will take shape? These questions are shaping the future trajectory of the industry.
What New Features Will Emerge from Integrating Large Language Models with GIS?
In natural language processing, OpenAI has popularized large language models (LLMs), but challenges persist: general LLMs struggle to adapt directly to GIS, while vertical-domain models often lack sufficient intelligence.
In June 2024, SuperMap GIS launched China's first geospatial intelligence agent application, SuperMap Copilot (Preview), based on AI Agent technology, alongside a suite of AI-driven industry knowledge generation tools.
By 2025, amid iterative advancements in LLMs, the rise of agent technologies, and accelerated open-source ecosystem development, what upgrades will SuperMap Copilot bring? How far has SuperMap progressed in independently developing and fine-tuning domain-specific LLMs? With the transformation from UI interactions into language-driven workflows, which products have realized automation and intelligence in handling complex tasks?
What Are the Latest Advances in Remote Sensing Interpretation Models?
The emergence of CV-domain models like SAM has opened new technical pathways for intelligent remote sensing interpretation. However, most general CV models are trained on everyday images, limiting their performance in remote sensing scenarios.
To address this, SuperMap released its self-developed Remote Sensing Interpretation Large Model (LIM) in June 2024. LIM is structured into upstream and downstream components: an upstream self-supervised learning network and a downstream classification task network. This architecture enables more effective learning of geospatial features in remote sensing imagery, achieving higher-precision object classification and recognition.
Against the backdrop of rapidly evolving self-supervised training techniques and vast unexploited remote sensing data, what enhancements can LIM deliver? How can large models adapt swiftly to diverse interpretation tasks?
How Is 3D GIS Innovating with AI Integration?
In 3D modeling, AI is driving revolutionary breakthroughs. Deep learning-powered intelligent modeling has redefined traditional paradigms, overcoming bottlenecks in efficiency, accuracy, and scenario adaptability.
For example, constructing 5,300 LOD 1.3-level buildings across 35 square kilometers in just 20 minutes has become routine—boosting efficiency by orders of magnitude.
Yet, this remains a shallow AI application focused on vector extraction and basic information recognition. How can we build more refined, realistic-textured architectural models and achieve efficient automated reconstruction of individual structures?
On June 24-25, all the answers will be revealed at the 2025 Geospatial Intelligence Software Technology Conference in China National Convention Center in Beijing!
The conference will focus on cutting-edge technologies and industry hotspots, showcasing the latest advancements in geospatial intelligence software technology and products, and exploring new opportunities for industry development. We look forward to your presence!
