×
Community Blog Frontier Technology | AI on the Cloud Helps Scientific Research

Frontier Technology | AI on the Cloud Helps Scientific Research

This article discusses Alibaba DAMO Academy's latest release, AI Earth.

By Alibaba DAMO Academy

Satellite remote sensing images are an important source of information for Earth observation. It can help people carry out disaster prevention and mitigation and protect Earth's resources more efficiently.

However, the link between obtaining, processing, and analyzing satellite remote sensing data was longer and more expensive in the past. Therefore, AI Earth (an earth science cloud platform that integrates satellite remote sensing data, remote sensing AI algorithms, cloud compute capacity, and cloud storage) can help scientific research work more conveniently.

Introduction

Recently, Alibaba DAMO Academy released AI Earth.

This platform integrates remote sensing data, remote sensing AI algorithms, cloud-based high-performance computing, and storage. It provides quantitative cloud AI computing and cloud-based storage for researchers free of charge. It helps researchers carry out analysis on scientific research work, such as agricultural disasters, climate changes, and water quality.

The starting point of the story can be traced back to half a century ago.

In 1972, the Club of Rome published a famous research report entitled The Limits to Growth. In this report, the most famous scholars in the world at that time deeply discussed the relationship between population, food, industry, pollution, and non-renewable natural resources and opened a new field of research.

Fifteen years later, the United Nations commissioned the World Commission on Environment and Development to prepare the first Our Common Future report, which summarized this research field as sustainable development. “Humanity can make development sustainable to ensure that it meets the needs of the present without compromising the ability of future generations to meet their own needs.”

1

After almost 40 years, at the 70th UN General Assembly, 193 UN member states adopted Transforming Our World: The 2030 Agenda for Sustainable Development, including 17 sustainable development goals and 169 specific goals. All humankind hopes to strengthen the understanding of society, economy, and environment and promote human society to embark on the road of sustainable development.

However, scientists quickly discovered that achieving such goals faces many severe challenges. For example, the scientific community does not understand the relationship between mutually restrictive goals. They also lack sufficient data and tools to improve cognition from these data-Algorithms that can process massive data in geoscience.

The ability to process data is one of the core challenges for the scientific community. In addition to the amazing scale, these data have spatial attributes, so they have big data traditional characteristics (such as massive, multi-source, heterogeneous, multi-phase, multi-scale, and strong temporal and spatial correlation, and physical correlation. Traditional algorithms and traditional computers are difficult to use here.

1
The Remote Sensing Image Data Retrieval Page of DAMO Academy AI Earth

DAMO Academy has also noticed these challenges and recognized that the high cost of remote sensing data acquisition, low degree of automation of existing analysis methods, high cost, and low efficiency of interpretation are hindering human's understanding of Earth, society, environment, and the interrelationships between them. Therefore, the AI Earth was launched.

AI Earth – What Can It Do?

AI Earth is a smart cloud service platform that provides one-stop Earth observations. It is built on Alibaba Cloud and can provide high-performance computing capabilities and storage services on the cloud.

The platform also integrates PB-level open-source satellite remote sensing data (covering the industry's most mainstream Landsat 8, Landsat 9, Sentinel-1, and Sentinel-2) and more than ten kinds of remote sensing AI algorithms (involving terrain classification, change detection, and SAR water extraction).

Among them, DAMO Academy has realized technological innovations in the remote sensing AI algorithm.

Based on NAS network structure search technology, a series of dedicated and efficient feature extraction network structures for remote sensing data are designed, such as the efficient classification network structure MuffNet.

In the aspect of target detection, a rotating target detection framework based on Transformer was proposed to achieve the best effect on DOTA and other remote sensing detection data sets.

2
AI Earth Terrain Classification Effect

Open Multiple Capabilities – Free for Researchers

Equipped with data, algorithm, computing, and storage capabilities, the platform provides researchers with the following capabilities to help improve the efficiency of satellite remote sensing data processing.

Multi-Source Data Retrieval: It integrates a PB-level of open-source remote sensing datasets. You can specify the search area and filter the conditions (such as the data collection time, data type, and cloud volume) to obtain the required data list. The retrieval results can be downloaded directly or processed in the cloud collection.

Online Data Processing: It releases the core capabilities of remote sensing AI of DAMO Academy and supports the simple use of twelve types of remote sensing AI online interpretation tools, including terrain classification, change detection, building extraction, parcel extraction, and SAR water extraction.

Cloud GIS Workspace: A project workspace is created from the perspective of professional GIS software to process raster and vector data online in a simple, convenient, and efficient manner.

Comprehensive Data Management: Users can upload raster or vector data independently and manage the collected public remote sensing data and the result data analyzed and processed by the platform. The data results can be directly applied on the cloud or downloaded to the local computer.

AI Model Training: The self-learning training feature of the remote sensing AI model will be launched soon for various business scenarios. Users can automatically train remote sensing AI algorithms in a visualized manner.

Developer Mode: Notebook-based developer mode will be supported later, which will be applied to remote sensing data processing and AI model training. It can provide exploration possibilities for more scientific research users and developers.

Weather AI Forecast Service: Based on the numerical model forecast technology, integrated with several innovative weather AI algorithms of DAMO Academy, the weather across the next ten days can be forecasted. It also provides three-dimensional forecasts at different heights and provides API interfaces for the entire industry.

3
DAMO Academy AI Earth Analysis Tool Template

Zhang Jingfa, a researcher at the National Institute of Natural Hazards of the Ministry of Emergency Management, said, “Earth observation research is in a new round of technological changes. The intelligent integration of cloud computing, AI technology, and remote sensing will promote the development of natural disaster prevention and other scientific fields.”

Summary

In September 2020, the AI Earth platform released by Alibaba DAMO Academy realized the fusion analysis of multi-source data, such as satellite images, autonomous aerial vehicle images, real-time video streams, meteorological data, and IoT data.

Li Hao, the Head of the AI Earth Team at DAMO Academy, said, “The task of remote sensing image analysis requires strong computing power and innovative algorithms. Our platform provides users with a good online research environment by combining the advantages of cloud computing and AI algorithms and uses cloud AI to help scientific research.”

0 0 0
Share on

Alibaba Cloud Community

473 posts | 24 followers

You may also like

Comments

Alibaba Cloud Community

473 posts | 24 followers

Related Products