Alibaba DAMO Academy releases a large model of remote sensing AI, allowing AI to further penetrate into the fields

On October 20, Alibaba DAMO Academy released the industry's first large-scale remote sensing AI model. A single model can identify farmland, crops, buildings and other surface objects, allowing AI to further penetrate into the fields and greatly improve disaster prevention and natural disaster prevention. The model has been opened for use on the AI ​​Earth geoscience cloud platform to improve the analysis efficiency of remote sensing applications such as resource management and agricultural yield estimation.

Remote sensing technology is widely used in the national economy and people's livelihood, such as urban operations, farmland protection, and emergency disaster relief. Remote sensing AI can greatly increase the depth of utilization of existing data and output more refined and accurate analysis results, such as Combining satellite photos and historical meteorological conditions, the growth status of crops in a certain farmland can be "calculated", so that farming is no longer passive, but more proactive, "depending on the weather."

In the past, due to the huge scale of remote sensing satellite image data and the complex classification of ground objects, to identify different surface objects, multiple dedicated remote sensing models needed to be trained separately, and a single model had low recognition accuracy and generalization. Problems such as sexual intercourse. In April 2023, the paper "Segment Anything" released by Meta brought computer vision into a moment of rapid iteration of large models, and also promoted the development of remote sensing AI in the direction of "one model solves multiple tasks".

The Remote Sensing AI Interpretation Universal Segmentation Model (AIE-SEG) proposed by DAMO Academy is the first to unify the tasks of image segmentation in the field of remote sensing. One model can quickly achieve "zero samples of everything" Extraction can identify nearly a hundred types of remote sensing surface objects such as farmland, water, and buildings, and it can still maintain high-precision recognition under multi-task processing, and can also automatically tune the recognition results based on the user's interactive feedback. In some specific scenarios, compared with traditional remote sensing models, the accuracy of instance extraction can be increased by 25%, and the accuracy of change detection can be increased by 30%.

Caption: This model supports multi-modal interaction. For example, if you enter "Extract farmland from the image", the selected target will be automatically identified

Based on the above The basic capabilities of the remote sensing AI large model provide "out-of-the-box" API calling services. Users can customize different remote sensing AI interpretation functions according to different needs, such as water extraction, farmland change monitoring, photovoltaic identification, etc.

The Shandong Provincial Institute of Land Surveying and Mapping has been cooperating with the Damo Institute in the fields of natural resources survey and cultivated land protection since 2022, using a large remote sensing AI model to monitor the growth of winter wheat in Shandong Province, with a recognition accuracy of 90% % or more, effectively improving the efficiency of winter wheat remote sensing interpretation, helping agricultural managers better predict grain yield and improve agricultural production efficiency.

The National Institute of Natural Disaster Prevention and Control uses a large remote sensing AI model to identify landslides and collapsed buildings. In the test of remote sensing images of historical natural disaster areas, it only takes more than ten minutes to extract this disaster information. It is dozens of times more efficient than manual identification methods, providing efficient and accurate remote sensing analysis support for scientific disaster relief.

Luo Hao, head of the AI ​​Earth algorithm of DAMO Academy’s Vision Technology Laboratory, said that remote sensing multi-modality is the only way to advance human beings to better understand the earth. DAMO Academy will continue to promote the research of large remote sensing AI models. Use AI to assist the exploration and application of earth science.

AI Earth is a one-stop earth science cloud platform released by DAMO Academy in 2022. Based on the accumulation of technologies such as deep learning, computer vision, and geospatial analysis, it provides cloud computing analysis services for multi-source observation data. Currently, it has established cooperation with 50 domestic universities, and related technologies have been applied to institutions such as the Ministry of Water Resources, the National Meteorological Center, and the Ministry of Ecology and Environment.

Attachment: Entrance to use the large remote sensing AI model of DAMO Academy

https://engine-aiearth.aliyun.com/#/app/aie-seg

The above is the detailed content of Alibaba DAMO Academy releases a large model of remote sensing AI, allowing AI to further penetrate into the fields. For more information, please follow other related articles on the PHP Chinese website!