Selected Readings
Since the second half of 2023, AI is undoubtedly the hottest technology topic, and it seems to be bringing us a whole new future world. AI, or Artificial Intelligence, is based on the underlying logic of enabling computers to have certain thinking capabilities through massive amounts of data and specific algorithms, which is also regarded as an enhancement and extension of the human brain.
While people are looking forward to the opportunities and changes that AI may bring, AI technology has been gradually applied to practice in the field of natural resources, and has achieved corresponding results. How to keep up with the pace of technological development and integrate AI into various natural resource management functions and application scenarios is a new topic that natural resource departments at all levels will face.
Keywords: #自然资源# #人工智能# #深度学习#
Number of characters: 5000 characters
Reading time: about 20 minutes
AI prospecting
Moving forward in groping
At the beginning of this year, news that AI technology had found a giant copper mine in Zambia caused a sensation in the geological industry, and although it quickly proved to be a miscommunication, the discussion of "artificial intelligence prospecting" has risen sharply in China.
In fact, in the field of geology, AI-related explorations and achievements have involved many aspects such as regional geological survey, intelligent mineral prediction, geological hazard investigation, and mineral exploration. The earliest breakthrough and successful application was intelligent geological mapping.
Li Fengdan, director of the Information Engineering Research Office of the Natural Resources Comprehensive Survey Command Center of the China Geological Survey, introduced that after years of exploration, the China Geological Survey has basically formed a technology and method for the whole process of geological mapping, deep learning-based mapping object recognition and positioning, and artificial intelligence geological map generation technology and method, which is synchronized with the mapping process The AI application process and modeling platform have been demonstrated and applied in the geological mapping projects of the three major rock types and orogenic belts, and good results have been obtained.
"AI has a strong ability and level of prediction of geological maps in blank areas, which can not only expose the geological bodies in shallow-covered areas, but also analyze and compare the areas that have changed before and after through iterative geological maps, so as to provide reference for the deployment of geological routes and the study of key geological problems, so as to improve the accuracy, research degree and work efficiency of geological mapping." Li Fengdan said.
In terms of prospecting, due to the need for rich geological, geophysical and geochemical data, it is still difficult to fully apply artificial intelligence. However, relevant experts believe that the use of artificial intelligence deep learning methods, especially in the mineral resources prediction link, is used to find resource enrichment areas, delineate target areas, and improve the success rate of prospecting, which still has optimistic prospects.
Lou Debo, a researcher at the Institute of Mineral Resources of the Chinese Academy of Geological Sciences, has been practicing the above aspects for many years, and his research team has conducted experiments on deep learning semantic segmentation methods on granitic pegmatite lithium deposits in a working area of the Altyn area. Experiments have preliminarily proved that this method has a significant effect on solving the problems of complex distribution of ground objects and difficult identification of small granitic pegmatite veins, and can be used as a new prospecting and exploration method for granitic pegmatite lithium deposits in the difficult-to-enter areas of low vegetation coverage on the mainland, and provide fast and effective technical support for the prospecting and investigation of lithium mines.
However, he said that the advantage of artificial intelligence lies in the regular intelligent processing of big data, and when applied to specific fine exploration, it is also necessary to break through the barriers brought by the complexity and individualization of different mineral deposits. Models trained on small-scale data may not be adapted to larger-scale regions. This is due to the extreme complexity and inhomogeneity of geological information. In his opinion, in order to improve the performance of deep learning, as far as pegmatite prospecting is concerned, the first is to further improve the structure of the model and improve the prediction performance of the network in details; Second, the newly discovered ore-bearing pegmatite are continuously supplemented as training samples in the prediction, and the generalization ability of the model is improved through further training and optimization, so that it can be predicted under more complex conditions.
Compared with the personalization of "artificial intelligence + mineral deposit exploration", Li Yongsheng, a researcher at the Development Research Center of the China Geological Survey and director of the Mineral Policy Research Office of the Mineral Exploration Technology Guidance Center of the Ministry of Natural Resources, introduced that "artificial intelligence +" is more universal in promoting intelligent mineral prediction and boosting the deep integration of a new generation of information technology and geological prospecting.
Taking the China Geological Survey's intelligent mineral prediction technology research and prospecting demonstration project as an example, the biggest feature of the project is "data-driven + knowledge-driven". In terms of data, the massive geological maps accumulated in the field of geology and mineral resources for a long time are fully applied, and the semantic and formal unity and vectorized expression of various geological elements such as geological evolution, deposit formation, and ore deposit distribution in the drawings are realized, so that they can be effectively integrated with geophysical, geochemical, and remote sensing data, and quantitative analysis and calculation are carried out. In terms of knowledge, based on the knowledge graph of the "trinity" prospecting and prediction theory deposit model of metallogenic geological body, metallogenic structure and metallogenic structure plane, and metallogenic characteristic markers, the vectorization of geological knowledge characteristics is realized through graph embedding, and the metallogenic analysis and prospecting prediction of data and knowledge integration are realized.
According to reports, the project has been based on CNN, ResNet, ViT, twin network and other artificial neural network architecture of prospecting prediction methods, developed an intelligent prospecting prediction system, established a geoscience big data intelligent mineral prediction method technology system, and in Gansu Bridge, Zhaishang and other gold ore exploration areas, Jilin Baishan polymetallic ore concentration area, Shaanxi Shiquan-Xunyang gold mine integrated exploration area, Yunnan Ludian lead-zinc-silver polymetallic ore integrated exploration area and other more than 20 typical mining areas to carry out intelligent mineral prediction pilot applications. Drilling verification has been carried out in Gansu Bridge, Beishan and other places, showing good prospecting potential.
AI Governance
"Hui Na Land" debuted
If geoscience research and prospecting breakthroughs are technology-intensive fields, then land planning and management reflect more complex social and economic characteristics. Land-related planning, remediation, transfer, supervision, real estate registration, as well as the laws, policies, economic models, etc., constitute a rather complex system, with a large number of data, indicators, policies and regulations, costs and benefits to consider, which also provides a broad space for the application of artificial intelligence.
At the first National Territorial Spatial Planning Annual Conference held in August 2023, Wu Zhiqiang, academician of the Chinese Academy of Engineering, proposed that to realize the digital and intelligent transformation of modern national spatial planning, there are five key elements: goal vision, path selection, development momentum, accurate assessment, and iterative optimization. Among them, the accuracy of the target vision determines the soul of the whole plan. The goal and vision of digital intelligence empowered spatial planning is to provide a variety of possibilities and multiple solutions for the comparison of multiple possibilities and solutions through machine learning of the historical laws of spatial development, and to provide a variety of possibilities and multiple solutions for the target selection of the production, life and ecological "living-living" system on the land space and above it, and at the same time capture people's needs on a large scale.
Academician Wu Zhiqiang believes that the path selection of digital intelligence empowered spatial planning is to provide multi-mode prediction schemes through AI methods, focusing on core planning issues such as land scale, spatial structure, and functional layout, to help decision-makers and planners foresee future trends and determine development paths. He also demonstrated the intelligent deduction of the spatial development of Quzhou, Deqing and other cities under the three development modes of strict scale control, law-based and strategic guidance.
At present, many experts and scholars in China believe that the rapid development of AI technology will bring significant changes to urban planning and land use. The application of AI technology in these fields can not only provide more efficient tools, but also innovate ideas and tap more potential for urban development. Through machine learning and data mining, AI can identify and analyze land use patterns and predict future trends. This helps planners better understand the current status of land use and future development positioning in different regions, as well as the needs of environmental protection and infrastructure construction, and provides a scientific basis for the formulation of urban development planning.
In fact, AI technology has been gradually applied in land management in some places. Not long ago, Yuhang District, Hangzhou City, Zhejiang Province, implanted an AI model in the natural resources smart trading service platform, allowing enterprises to shift from "knowing how to take land" to "taking land wisely". The platform's AI intelligent matching model can help enterprises customize personalized land selection plans and obtain suitable land plots from thousands of land parcels.
According to reports, this AI model can form user portraits according to the user's browsing history, search preferences, etc., and the platform will recommend the land resources with the highest fit with the enterprise according to the user's demand characteristics. So far, 1.42 million people have used the service.
In addition, the AI customer service's full-process assistance in land acquisition can provide enterprises with 24×7 hours of "à la carte" service, with high response and accuracy in answering questions. For the land plots that the bidder has followed or registered, the platform will take the initiative to push reminder SMS, including key nodes such as resource release, registration completion, qualification acquisition, and transaction auction.
According to the person in charge of the Zhejiang Provincial Natural Resources Online Trading Center, the business environment in the region will be further optimized by mining the value of transaction data. "For example, if a company wants to know the requirements of the per mu tax on industrial land in a certain area, or wants to know whether there is an industrial land in a certain area recently, the AI model can quickly give accurate answers." The person in charge said.
It is understood that Zhejiang also plans to build an upgraded version of the AI model, use the big data of natural resource transactions, explore the overall allocation of all elements of natural resources, and provide support for government decision-making and enterprise development.
In terms of cultivated land protection, AI technology has also been applied to a certain extent. Cultivated land is the foundation of ensuring national food security, and the rapid and accurate extraction of cultivated land plot information is an important basis for supervision, and it can also provide technical support for agricultural yield estimation, disaster analysis, land consolidation, etc.
Through the combination of AI and high-resolution remote sensing, the functions of ground object extraction and change detection can be realized, and the extraction effect and accuracy have been greatly improved compared with traditional methods. At present, some domestic technology companies have built sample databases and developed relevant models by collecting the characteristic data information of various types of cultivated land such as paddy fields, dry fields, and terraced fields in different landforms. Some of them use the combination of surface and line extraction models, which can not only extract the range and obvious boundary of cultivated land, but also display the texture details inside the cultivated land, and obtain the final cultivated land plot by fusing the surface and line results