Insights From Above: Rethink Land With AI | Kehan Zhou | TEDxBoston

abroad to teach computers how to understand subtle images and generate important

insights

that help us plan and build more effectively in the physical world, so our journey began trying to find a way to quantitatively measure rural real estate after 2020, the urbanized became really obsessed with the rural. houses and

land

because these properties really offer you much more access to green spaces and more affordable prices and one of the most important steps when buying a rural house is to find out what is actually on the

land

, this is because the land coverage and The vegetation will directly affect your experience of living in the property and also from an investment perspective it can affect the value of your property but it turns out that understanding the land cover is not very easy, these properties are very large so it is physically impossible to walk through them and manually inspect the property As a result, many buyers and businesses turn to frame images for information.

Well this can work to some extent, there is still a missing method for two important reasons, number one, just looking at subtle images doesn't really give you an idea. A good understanding of what the property has depends too much on the person's perception and intuition to provide quantitative data that allows you to compare similar properties objectively and the second thing is to review subtle images, it can be a very tedious task and is practically impossible. To do it on a large scale, for example, if you want to analyze the state of California, you have to go through almost a million images, so we thought, hey, why don't we teach computers to understand subtle images and use this information to inform us ? what they see, after all, computers don't really get tired and we don't really have to pay them well, that's only technically true, so that's exactly what we did: we applied computer vision to subtle images and the result was really surprisingly reliable and accurate, so on the left side you can see the original raw image and on the right how the computer perceives it.

This not only saves us a ton of time but also gives us really important data that we never had access to before. The first type of data is a quantitative way of measuring different types of land covers, for example you can see that the AI ​​not only maps different types of vegetation and terrain but also the relative size of each and with this data we have a way very powerful way to quantitatively measure our land and resources. we can use it to measure the yield of a farm, measure the severity of a drought and even the buildable areas for empty plots and that's not all subtle images and applying computer vision to subtle images gave us really important

insights

by giving us spatial relationships between different types of objects, so here you are looking at two images of swimming pools, one of them is an above ground pool and the other is an inground pool.

It turns out that in-ground pools contribute more to your property value than buffer pools and computer vision. Tools can tell the difference between the two by looking at the space immediately surrounding the water and it turns out that in-ground pools tend to be surrounded by paved areas, while above ground pools tend to be surrounded by grass or trees and using this technique We can detect houses that are two trees closer that could be vulnerable to forest fires. We can also estimate the flood risk for coastal structures. Applying computer vision to the physical world gave us a way to digitize our planet in an indexable and searchable way.

It creates a scalable way to understand the information and land we have and you can also configure the algorithm to look for particular combinations of terrain patterns and structures and if you apply the same technology to the same region over time you will get additional insight into how a region evolves. an app we built for local government to detect ownership changes over time. Municipalities worry about ownership changes because they directly affect the value of the property and therefore their tax revenue, but in rural regions they must manually drive around and look for these changes that may or may not exist.

Now it is feasible, but with computer vision we can monitor a very large region, detect changes and alert authorities accordingly, so 2023 I think will be a very exciting year for computer vision, we have more access to images subtle and algorithms are becoming smarter and smarter every day will change the way we deal with the impact of climate change mitigate natural disasters plan our crops more effectively and even assess damage to get more. Today I will leave you with this thought if you are currently a consumer of space technology. subtle data or images, think about how to incorporate computer vision into your workflow to make your organization more effective and, if you are in the rural property market, you should also try using computer vision to know your customer potential and make more informed decisions. you are very foreign