Journalism, AI and satellite imagery: how to get started

Satellite image of the Amazon. Tocantins, Brazil. Source: Copernicus Vigil data [2022] processed by Vigil Hub, by Highlight Optimized Natural Color.

In the first of two guest posts for the OJB, first published in ML Satellite, the MA student newspaper Federico Acosta Rainis explains how to get started with satellite journalism, and avoiding common pitfalls.

Working with satellite images and AI models takes time and patience. There is no general rule: you have to find the right model in each case, in a process of trial and error, when gathering large amounts of data.

Here is the plan Anatoly Bondarenkodata editor Texty, it depends;

“Find a business or a story where you don’t regret many resources.”

The combination of satellite images and AI is useful for obtaining a general context where no data exists, finding patterns over large territories or time, counting scattered objects or finding a needle in a haystack.

“AI makes sense when you want to do things more often or more often,” he says. Edward Boyda e* Earthrisewho in the Amazon Mining Watch project. “Either because you are working in such a large space that it is impractical for a person to do it or because you want to do it again and again.”

Before you go any further, ask yourself: is this the best tool to tell a story?

Perhaps there is an easier or more efficient way to get the information you are looking for. The dataset may already be available, or you can obtain it through a public information request. Or manually count objects in a satellite image, without the need for AI.

Know the terrain and basic tools

Amazon Mining Watch website
Amazon Mining Watch website

If you want to go ahead, the next step is to get used to yours region of interest (ROI) and the tools you use.

One of the advantages of satellite data is that it allows you to explore from afar: you can observe any corner of the globe from your computer. But you still need to get used to ROI to properly interpret your findings.

If you can attend, so much the better; if not, collect as much as you can.

Find out what are the problems or hot topics, non-satellite images, contact local people or experts who work there. This will help you to reduce errors in the analysis.

Boyda says that satellite imagery is useful for investigating areas where “it’s difficult for journalists to work,” but that it should always be used “in conjunction with traditional reports on the ground,” as satellites are “useful to illuminate or quantify; but I don’t bring that element into human history.” .

Also Read :  Workforce development is a priority for MVCC’s AIM

As for the tools, take some time to experiment and choose the ones that are most comfortable for you or that you can best adapt to your situation.

Starting with a note: if you already know R or * Pythonfor example, look for libraries such as RSToolBox or Earthpy.

You don’t always need to write a code: Free software with AI algorithms to analyze images or detect objects. Two very useful ones are QGIS (with SCP and dzetsaka plugins) and SNAP (European Space Satellite sample data).

For the initial analysis, you need not to like anything too much:

  1. Use Google Earth Pro to quickly see ROI over time.
  2. If you want more information, the Police Hub’s EO Feed is good for a first analytical approach: you can use different satellites, test band combinations to spread the cloud cover.
  3. If this is still not enough, it’s time to install your device by combining different tools.

Ask for help from others

Most journalism pieces that use AI and satellite imagery collaborative projects and is trusted by a given expert.

You can seek help in specialized forums or in contact with other journalists who have used such techniques.

A map showing the earth's satellites in different intervals and categories: science, Copernicus and meteorology
Earth observation missions: image provided by the European Space Agency (ESA); (CC BY-SA 3.0 IGO)

But they also look beyond journalism; Earth Observation (EO) Remote sensing is used in many fields such as agriculture, ecology, biology and disaster management.

If you ask for advice from others before you start technical work, you will avoid some dead ends that others have already gone down.

It’s not just about solving technical problems. Whether an NGO working in environmental matters can provide high-resolution images, it is difficult for you to provide information about your ROI or put you in touch with local experts.

“This knowledge is not something intrinsic to the news industry,” he said Mathias Felipe de Lima Santosmanager of data-driven projects at InfoAmazonia and researcher at Digital Media and Observatory at the Federal University of São Paulo (Unifesp). “So news agencies need to go where they can find, looking at civil society or startups using satellite images.”

In this way, he says, institutions can “carry out research that goes beyond their boundaries.”

Start small and get organized

IT ALWAYS STARTS SMALL. Even if your final ROI area is much larger, start by limiting your analysis to the area that is representative of the phenomenon you want to investigate, but as small as possible.

Also Read :  NVIDIA Teams With Microsoft to Build Massive Cloud AI

The use of AI in one satellite image captures a fair amount intellectual effort, computational power and Processing Time.

When you add more images, the difficulty increases exponentially: the data to be learned increases and new problems arise, such as stitching images together, working out overlaps, comparing images from different periods.

Do all the testing you need in a small area and only when you are satisfied with the results extend the model to your entire ROI.

You may use a lot of information from different sources. The satellite image in its raw form often contains complex names, combinations of letters and numbers that refer to the coordinates, date and type of processing used.

It’s good to be consistent: using a file as a master file can help you organize all this data.

As in written code, it is better to use extra-explicit nomenclature than generic names, which make it difficult to remember what they mean.

Working with AI: models and computational power

Semantic segmentation involves distinguishing different elements in an image. Image by B.Palac

AI comes in different flavors, but you’ll probably use one (or a combination) of the following tools;

If you’re thinking of using a model of object detection, check to see if they already exist image datasets It is available for installation. Creating datasets is time consuming.

Try different models – and different parameters – to see what you like best. Some are well known Random Forest, Supporting Vector Machine (SVM) and Close neighbors.

Multi-layered Deep Learning (DL) models tend to give better results, but are much more complicated to build.

If you give a lot to something, consider it cloud computing. There are free tools such as Google Colab or Google Earth Engine (GEE) that can break down huge amounts of data and get them in a short time.

GEE also provides easy access to datasets from Landsat, Vigiland WAYS embassies

Amazon Mining Watch, for example, uses Cartesian Labs, a free cloud computing service to process large amounts of satellite data.

“It’s built around so you can break images into tiles for parallel processing and feed them directly through the model,” says Boyda. “So we don’t have to think about that side of the pipe.”

If you work on your personal computer, you need a solid-state disk (SDD), at least 8 or 16 GB of RAM and a good amount of storage: one raw satellite image often weighs several gigabytes.

Also Read :  She built a startup that raised millions, shares 3 tips

A powerful processor, a dedicated video card and a good-sized monitor will also make your work easier.

Identify the results and share your method of delivery

Skepticism is a powerful weapon in journalism. Treat AI predictions like any other data: don’t take them for granted.

Like all technologies, AI does not work magic and does not replace traditional journalistic methods. but he completes them.

“This is always the golden rule: 80% of the work is given,” says Anatoly TextyInvestigation into illegal amber mining in Ukraine. “Also, the correct interpretation of the data, because we had to find images in high resolution, but no less we had to find some important examples of mining patterns.

It will be a good part of your job to clean up false positives or why so many false negatives to improve your model.

If you’re working on a story about change, for example, pay attention to what hasn’t changed, and ask yourself why.

It is important to find some ground truth of the results of your model: given the field directly in the area that the analysis has obtained: is something similar to the crop field in the satellite image, really the crop field?

That doesn’t mean you have to travel: you can reach out to local experts to evaluate your findings.

When you preach, share the methodology whether you have placed it in the same or in an external place GitHub. It is a good use of transparency, it helps to understand your analysis and it helps others who are working on the same topic.

But also think about it audiencewho do not necessarily know how AI works, and explain its scope and limitations.

“Journalists need to be transparent and not only explain metadata, but also explain what AI is, what computer vision is, and how we approach it,” says Mathias. “And make the limitations clear” [of the model].

We will always explain the gaps in your research, the things that have not been discovered and the level of uncertainty in your model. Remembering that you could not pay does not diminish your work; nay, it makes more certain.

Federico is a Google News Initiative Partner at The Guardian. You can find him on Twitter @facostarainis

Advertisement

Privacy settings



Source

Leave a Reply

Your email address will not be published.

Related Articles

Back to top button