Picterra, a Swiss AI-based SaaS platform allows users to interactively create a personalized AI detecting, localizing and counting any objects from satellite and aerial imagery. The company aims to democratize geospatial mapping, and its platform bridges the gap between Earth Observation (EO) imagery, cloud processing and geospatial insights by commoditizing Machine Learning technology. From precision agriculture to utilities and infrastructure, Picterra serves a wide variety of clients and provides customized services.
Its main partners are geospatial and UAV mapping professionals looking to derive insights and actionable information for specific verticals based off large or heavy EO imagery set. The Picterra platform allows users to seamlessly integrate cutting edge machine learning technology into their existing workflow, so they can focus on their core business while achieving quick return on investment.
The company offer users – developers of vertical applications, GIS integrators or geospatial services providers – a tool to build, integrate or deploy their own detector at scale. While using Picterra, users create a strategic asset for their own business, like a library of valuable trained detectors and extend their portfolio with new applications.
Picterra has also created a create a community driven library of deep learning algorithms that works “transfer learning,” a research axe in ML in which knowledge acquired by a model when recognizing an object to speed up the learning of another model is used.
“Our platform could spark the development of new geospatial-based applications we can’t even imagine yet – similar to how ubiquitous GPS led to companies like Uber and Airbnb”, says Pierrick Poulenas, CEO and co-founder, Picterra, in an exclusive interview.
Picterra enables users to extract satellite and drone imaging insights through a combination of AI, Deep Learning and other algorithms. What are the main advantages of this approach, and do you think AI would revolutionize image analytics?
Everyday terabytes of EO imagery are collected through satellites, drones and other observation platforms. They are increasingly able to pick up high spatial and temporal images, so access to EO imagery is not the main blocker anymore.
But as we just touched on, the main challenge is the ability to pull out useful facts and figures from massive datasets – without having to pore over maps and count objects individually or invest in expensive and time-consuming data science programs. That’s why machine learning is certainly the only viable option to extract at scale meaningful and valuable information while creating an extremely valuable asset: a library of trained models. Support from organizations like Omidyar Network allow us to help solve this challenge, to the benefit of the entire industry.
How can users build their own AI detector using Picterra and train an algorithmic model that will provide pertinent insights?
With our platform anyone –not only data scientists – can train and deploy at scale (from one hectare field to an entire region) features detectors.
The whole process takes about 10 minutes. Users simply draw a few polygons (annotations) around the object or features of interest, then Picterra’s platform uses these annotations to learn how to detect them. The user has then created a “detector” they can reuse on another set of images or improve with new annotations to fine-tune and increase accuracy.
With Amazon launching Groundstation-as-a-Service, do you think AI as a service for Earth Observation will soon become widely popular?
AI-as-a-Service has the potential to deeply impact the geospatial industry by allowing a faster and more accurate extraction of meaningful information from an ever-growing amount of EO data. We’ve already seen it being used for everything from mapping informal settlements and building construction to tracking storm damage and counting animals, trees, and cropland.
I also believe the impact of AI-as-a-Service will go beyond the EO industry and impact every mainstream industry.
How do you think technological convergence and dedicated launch options for small satellites will impact the Earth observation sector?
It will first reduce the cost of the EO data themselves and therefore enabling new applications that were previously not economically viable. Small satellites, as well as constellation satellites, will also trigger the development of real-time imagery-based applications that big, monolithic satellites were not able to provide. However, this new, real-time and high-resolution data will require a new paradigm of the way we build and even imagine digital infrastructures to make sure the insights extracted are shared as widely as possible, bringing society a new source of knowledge.