Ai+ remote sensing: releasing the value of each pixel

The theme of this speech is ：AI+ remote sensing , Release the value of each pixel . It is mainly divided into four parts ：

• Remote sensing big data era

• AI How to release the value of remote sensing pixels

• Main origin products and typical applications

• Summary and prospect

Remote sensing big data era

remote sensing , That is, distant perception .1839 year , After the first camera came out , People try to put various imaging devices on a higher platform , Look at the world around us from a broader perspective .

WAVE SUMMIT+2021 Deep learning developer Summit

【 Cross front ,AI Co extension 】 Forum

At present, the world has entered the era of hourly rapid response and sub meter remote sensing observation big data , Remote sensing takes electromagnetic wave as information carrier , Greatly expand people's perception ability . While remote sensing expands to a wider range of perception , It also brings broader application prospects .

From the perspective of remote sensing observation ability , From the 100 meter scale in the early years to the current meter scale 、 Sub meter level 、 Centimeter scale observation capability , The increase in the amount of data leads to the expansion of application prospects , But it also brings unprecedented challenges to interpretation , Traditional manual interpretation is difficult to meet the needs of the rapid growth of data , So that it seems helpless in front of Remote Sensing Applications .

The rapid and automatic extraction of remote sensing information has become the bottleneck of the whole remote sensing industry chain , Artificial intelligence is the latest and most epoch-making technology in recent years , Deep learning has made a breakthrough in many fields such as computer vision , As a special image, remote sensing image is bound to enjoy the dividend of this technology . however , How to make AI And remote sensing data to release pixel value , This is an issue worthy of discussion .

AI How to release the value of remote sensing pixels

1. remote sensing AI Where does the ration come from ？

Thanks in large part to the huge image library , Compared with natural images, remote sensing sample database is very few , The number of samples is also very small . This is mainly because the remote sensing sample database is not only a scene 、 The imaging process is complex , Moreover, the formation of sample database is also a very complex process . After ten years of Engineering iteration, our team , A large-scale remote sensing sample database has been established , Effectively cover the whole country / Partial global , It can meet the R & D and production of more than 50 kinds of products .

2. With data , How to input image data into AI In the system , That is to solve the problem of eating .

The deep learning framework is the solution to this work , But in the face of multi-channel 、 Big picture 、 Remote sensing data with geographical coordinates and rich information , The existing basic framework of deep learning and the framework for natural images can not be well applied to remote sensing data . therefore , Based on our understanding of remote sensing data and in AI Experience in intelligent analysis , After several years of engineering research and development , The first set of special deep learning framework of full remote sensing system in the industry is studied , It can meet the requirements of remote sensing segmentation and classification 、 target recognition 、 Change detection and remote sensing parameter inversion .

3. Deep network model how to digest and absorb remote sensing data , Fully extracting remote sensing information is the key to releasing the value of remote sensing .

Remote sensing has rich spectrum 、 radiation 、 Texture and semantic information , Among them, the of remote sensing images “ Spectrum ” Feature is the most essential difference between remote sensing image and natural image . But how to combine the geoscience scene 、 Build a deep learning network algorithm model integrating remote sensing features , This is the most critical link to release the value of remote sensing . For example, the cultivated land here is green under the visible light and shadow , The use of multispectral images can distinguish different crops , That is, more accurate classification can be realized by using the spectral information of ground objects .

4. How to improve model generalization ability , Realize the application requirements of large scenarios , This is a problem that must be solved in the development of remote sensing applications .

Remote sensing application scenarios are huge and very complex , Then superimpose the complexity of remote sensing imaging process , bring “ Same thing different spectrum 、 Same spectrum foreign matter ” It's very common . Take the commonly used water body as an example , Differences in sediment and chlorophyll in water 、 Different water quality 、 Under different temperature and light conditions, the water body is greatly different ; Another example is the seemingly simple outdoor track and field field, which is not only different in itself , And there's a lot of “ Li Kui ” To make trouble , The superposition of many factors makes it very difficult to realize accurate identification in a large range . On the one hand, these problems depend on our large-scale sample database , On the other hand, it also depends on the improvement of model algorithm .

5. How to simplify or even automate the process , This is the premise of remote sensing to ordinary people's homes .

So , We developed the spirit AI ImageBot Intelligent simple platform , The whole process from the original satellite image to the generation of remote sensing thematic information products is constructed 、 automation 、 Flow rapid product line , It can realize the of remote sensing information products “ Near real time ” production 、 Algorithms and products “ Evolution and iteration ”.

Let me show you two videos , It is a part of our system , The video on the left is an experiment for the Ministry of land and resources . Users only need to open the web page to extract , Do its dynamic monitoring on the basis of information extraction , Extract change information and identify key elements , Further inversion of its production capacity 、 State, etc . The identification on the right is to the road 、 Typical man-made features such as buildings are monitored , In a sense, this information reflects the most essential monitoring of remote sensing , One is to find out the home , The other is dynamic detection .

The following video is a concentrated display of products made nationwide and even globally in the process of R & D in recent years , This video is also CCTV News channels and evening news programs have reported , Up to now, we have identified more than 40 Chinese targets and geographical elements in all aspects . First of all, these elements cover all remote sensing target recognition information extraction at the technical level 、 Full technical chain and full technical difficulty coverage . Secondly, the superposition of these data provides us with another way to understand and recognize the world , For example, the video shows the classification results of the national two meter satellite images , Through the results, we can see the distribution of buildings and roads in the country and how to reflect the relationship between natural geography and human development , At the same time, we can also provide natural resources for the whole country 、 Ecological environment monitoring provides high-precision monitoring data .

Main origin products and typical applications

Previously mentioned is the information obtained from remote sensing images , In practice, remote sensing technology AI Very widely used . for example ： These natural resources listed 、 Urban supervision 、 Global strategies have broad application prospects , Can make their own products for different industries .

1. This picture is the first time that we have realized the full-automatic recognition of national targets by using meter level images , We use a server in 3 The national outdoor playground identification was completed within days , The accuracy is more than 95%, It should also be the first time for the industry to realize national product production at such high resolution .

2. Based on these thematic data on a national scale , Even simple color rendering can transform China's economy 、 The distribution of population development is clearly outlined , This is the charm of big data .

3. Comprehensive utilization of multiple information , Through spatial distribution information and time monitoring information , Be able to explore many internal laws , For example, the analysis of the causes of haze in Beijing, Tianjin and Hebei , Obviously, industrial pollution is the main culprit of smog .

Sum up , Remote sensing has recorded the real development process of mankind for decades , We use this AI+ Remote sensing data can improve the ability of rapid analysis , Realize the monitoring of changes in information over the past 40 years . We can also observe and think from a global perspective . For example, put global steel mills 、 Thermal power plants 、 Put the cement plant together , From a global perspective , There are four regions , Two European and American developed countries , Two developing countries —— India and China ; In terms of specific data , China accounts for half of the country , This shows that national development mainly consists of GDP The total amount and growth rate determine . Through these data , We can carry out many applications .

Summary and prospect

1. Just to summarize ,AI+ How to realize the application subversion of remote sensing data ？

From this picture we can compare , Compared with the traditional physical model , The production mode of deep learning is not only simple in the production process , And the cost is getting lower and lower , The accuracy is getting higher and higher , This is also our goal .

2. Here are two production pictures , I believe you will be touched by this picture , The combination of remote sensing and artificial intelligence , Can achieve a larger range 、 More abundant 、 More detailed information of various surface space-time elements , This information can provide enabling possibilities for a wide range of application industries .

Last , I want to say AI+ Remote sensing empowerment is a golden key , I hope everyone can have such a golden key .

Related to recommend

• AI+ Supercalculation ：AI Enable scientific and Engineering Computing Based on grid discretization

• AI+CFD： A new method and paradigm of scientific machine learning for aerospace dynamics

• AI+ Drug design ：AI Pharmaceutical and industry frontier
  

Official account , Get more technical content ~

This article is shared in Blog “ Flying propeller PaddlePaddle”（CSDN）.
If there is any infringement , Please contact the [email protected] Delete .
Participation of this paper “OSC Source creation plan ”, You are welcome to join us , share .