Recently, the second phase of the practical training course for surveying and mapping professionals in the municipal planning resource system was held as scheduled. This training was conducted in a combination of on-site visits and special lectures, with the theme of "The Development Road and Industrial Application of Artificial Intelligence Large Model". 45 core trainees from surveying and mapping related majors from the Municipal Planning and Resources Bureau, its subordinate units, the District Planning and Resources Bureau and the relevant departments of the dispatched agencies went to SenseTime to participate in the on-site training, and other personnel participated in the event through online live broadcast.
Field visits
A promotional video about artificial intelligence kicked off the event. Ms. Zhang Lin, General Manager of the Intelligent Remote Sensing and Smart Agriculture Division of SenseTime, introduced the development history, infrastructure construction and the company's safeguard measures in terms of integrated computing power to the participants in detail. Combined with the on-site display screen, interactive experience devices and product application scenarios, it comprehensively demonstrated how SenseTime applies artificial intelligence technology to urban operation management, remote sensing interpretation, autonomous driving, smart healthcare, smart home, augmented reality and social metaverse and other cutting-edge fields. They had in-depth exchanges on professional and technical fields such as intelligent remote sensing interpretation, automatic modeling, and digital twins.
Thematic lectures
Zhao Rigetu, director of the Intelligent Industry Research Institute of SenseTime Technology Co., Ltd., gave a special lecture and introduced the development road and industrial application of artificial intelligence large models to the trainees of the training course.
Since the Dartmouth Conference in 1956, artificial intelligence technology has experienced ups and downs, and the breakthrough of deep learning after 2012 has brought about a revolution in AI+ scenario applications. In recent years, large models represented by ChatGPT have led the second revolution of artificial general intelligence (AGI), solving open-ended tasks in an efficient way, and promoting a new research paradigm, that is, based on multimodal base models, through reinforcement learning and human feedback, constantly unlocking new capabilities.
At present, large-scale model technology has become a global hot spot, and international technology giants and domestic forces are actively participating in large-scale model technology competitions to promote the surge of computing power, technology acceleration, and application change. Computing infrastructure is characterized by a surge in scale, an increase in supply, and a decrease in costs, but the relationship between supply and demand is still unbalanced. The large model technology is developing in two directions: larger parameter quantity and suitable parameter quantity, and the modal evolution and application methods are constantly enriched.
Artificial intelligence applications are on the eve of change, and large model tools have shown potential in improving work efficiency and human-computer interaction, heralding the arrival of a revolution in productivity applications. At the same time, the development of large models also involves non-technical factors such as policy environment, science and technology competition, and open source ecology, and the global AI competition has become a competition at the national level.
In the future, the development of large models will expand to more modalities, applications and edge devices, the depth of technology will touch the human-like level, and in terms of basic innovation, researchers are exploring more efficient architecture, learning ability, self-training and self-feedback ability. This shows that large model technology will continue to evolve and promote the development of artificial intelligence to a higher level.
Communicate and interact
Qiao Zhenmin, Director of Natural Resources Industry of Intelligent Remote Sensing and Smart Agriculture Division of SenseTime, introduced the remote sensing large model & NeRF 3D reconstruction technology in combination with the professional field of surveying, mapping and geographic information.
With the rapid development of mainland aerospace technology, the ability to obtain satellite remote sensing data has been significantly enhanced, which is of great value to many industries such as natural resources, meteorology, and agriculture. However, there are deficiencies in remote sensing technology in information perception and application analysis, such as the mismatch between traditional operation methods and business needs, which makes it difficult to give full play to the value of remote sensing data. In order to solve these problems, intelligent remote sensing information extraction technology came into being, which has gone through three development stages: small model, large parameter model and multi-modal large model. At present, the industry is in an exploratory period of transition to multi-modal large models, which improve the understanding ability of remote sensing images through multi-modal information input and unsupervised methods. The intelligent remote sensing software based on the remote sensing model is mainly divided into general analysis software, model training software and intelligent remote sensing cloud platform, which is widely used in provincial natural resources monitoring and supervision, and has been used by relevant departments of more than 20 provinces and cities to assist in the work of intelligent remote sensing change detection and information extraction technology.
Remote sensing is a means of data collection, through data collection and data perception, the use of real 3D can have a better display and visualization of the collection results. NeRF 3D reconstruction technology can provide important support for the construction of real-life 3D in China, and can realize high-precision 3D reconstruction of city-level large-scale space and restore the real scene. Combined with AI model editing tools, it can simulate different weather scenarios, measure spatial data, and realize element segmentation, addition or removal. NeRF technology is combined with remote sensing large models to provide a comprehensive technical solution for the construction of urban 3D scenes and help the construction of urban digital twins.
Thinking: The integration of artificial intelligence and surveying and mapping geographic information
Artificial intelligence is regarded as a symbol of the fourth industrial revolution, and developed countries and technology companies around the world are actively investing in R&D and layout. China is also trying to build a first-mover advantage in the development of artificial intelligence, and promote the upgrading of AI technology from perceptual intelligence to cognitive intelligence. As an important technical approach to the current AI development, human-machine hybrid intelligence emphasizes application first and application-driven technology development.
The artificial intelligence innovation ecosystem in mainland China has been initially formed, and it is in the first echelon of the global AI index ranking. Major domestic technology companies such as Baidu, Alibaba, Tencent, iFLYTEK, SenseTime and Huawei have made significant progress and created open platforms in the fields of unmanned driving, urban brain, intelligent medical care, speech recognition, image and video processing.
The combination of AI technology and surveying and mapping geographic information is both a challenge and an opportunity. After experiencing digital transformation, the surveying, mapping and geographic information industry is facing the need for technological upgrading, especially in real-time data acquisition, information automation processing, and knowledge-based services. AI has had a wide impact on the discipline of surveying and mapping, promoted the theoretical infrastructure of intelligent surveying and mapping, and promoted the transformation from digital surveying and mapping to intelligent surveying and mapping.
Intelligent surveying and mapping takes knowledge and algorithms as the core, and builds a hybrid intelligent computing paradigm to solve high-dimensional and nonlinear problems that are difficult to cope with traditional surveying and mapping. It aims to automate simple, repetitive, and dangerous surveying and mapping tasks, unleash human ability to create knowledge and think spatially, and improve surveying and mapping productivity. The implementation path includes the analysis and modeling of natural intelligence in surveying and mapping, the construction of hybrid computing mode, the research and formulation of intelligent mechanism and computing mode, and the establishment of intelligent surveying and mapping application system.
The application scenarios of intelligent surveying and mapping are very wide, including new basic surveying and mapping, real-world 3D, digital twins, BIM, CIM, spatiotemporal knowledge services, etc. The application hotspots on the market side also reflect the industrialization value and sustainable growth potential of intelligent surveying and mapping. The future development of intelligent surveying and mapping requires deeper technology integration, wider application expansion, and higher level of intelligent technology to achieve richer data, information and knowledge products, and promote the transformation and upgrading of the surveying, mapping and geographic information industry.
(Source: Municipal Institute of Surveying and Mapping)