Mapflow.ai đ„ â a new application for automated mapping using satellite imagery
- Data processing platform
- Platform use cases
- Semantic analysis models
- Data sources
- Getting started with Mapflow
- References
Traditional digital mapping requires an impressive amount of work and a private army of cartographers to map even a single city. The process is pretty straightforward but the time-consuming work makes it costly. Hence we see the growing interest in the idea of AI-assisted mapping, powered by deep learning image recognition algorithms. We call it âlazyâ mapping, instead of using the buzzword âsmartâ (check out âSmart Mappingâ â some big companies have already taken over this term) which is used in relation to any systems that use artificial intelligence. On the other hand, what should automation systems do, but free us from routine work and help us speed up the creative process?
Our new product Mapflow.ai is not about substituting a productive cartographer with a lazy one, who delegates all his work to a machine. Mapflow is about how to use âlazy mappingâ in a smart way. đ€ đ€ đ€
Data processing platform
In Geoalert, we use CNN architectures pre-trained on cats and dogs đ¶, that is in ordinary photos, but we train them on large cartographic datasets â thousands and hundreds of thousands of objects. This is whatâs called deep learning.
While we were struggling with the issue of processing hundreds of millions of satellite imagery tiles, we implemented a prototype of the platform and launched a project of automatic (lazy) mapping all buildings over Russia. Next, we started to work on a new product.
While designing user scenarios, we assumed that digital cartography can be represented using three main stages of data processing:
- Access to Earth Observation data (for example, satellite or aerial imagery)
- Automatic preprocessing and analysis (welcome Mapflow!)
- GIS-facilitated validation and interpretation (verify automatic results and prepare them for further work)
Having summarizing these steps, we designed an application that:
- Has access to global satellite imagery data and can fetch data forany location in the world.
- Provides a single dashboard for managing AI-assisted mapping flows
- Allows you to run post-processings by adding new blocks related to the appropriate âmapping flowâ â like polygonization, merging with Openstreetmap, etc. for âbuildings detectionâ
Has external interfaces. Primarily itâs a REST API, secondly itâs some external apps like geojson.io that allow to visualise results.
Platform use cases
Most of our users are companies that use our API to integrate Mapflow into their own workflows or applications. API allows to create projects, start processings, get status, and stream processing results in GeoJSON format.
Use case #1 â Vegetation monitoring in powerline corridors
In this joint project with Skoltech Space Center, we developed a QGIS plugin for detecting and classifying forest vegetation in satellite imagery . It provides the cartographers at the Russian Powerlines Technical Inspection with data processing and report generating tool inside the work environment of their choice.
This QGIS plugin is interacting with Mapflow API to connect to the global satellite images by Maxar SecureWatch and generate mapping alerts with vegetation reports (the area covered by vegetation, the average heights of the vegetation).
We are telling about this product in this post.
Case â extraction of road masks for creating road network graph
Our partners from Geocenter-Consulting use Mapflow API to launch road extraction in satellite imagery, then import the road mask into their application âRuMapâ for further postprocessing and creation of the linked road network graph.
From the vectorized polygons of the road mask, the Geocenter âRuMapâ builds the skeleton of the road network, which is used to shape up the road graph.
The release of the product âRuMap: RoadNetworkBuilderâ is scheduled for February 2021, we will definitely tell you more about it along with our partners.
Case â country-wide processing of âbuilding footprintsâ
The project in which Mapflow scaling and batch-processing technologies were applied first was the automatic mapping of buildings over Russia. (Read here)
In the next country-wide-scale projects, our partners Kontur Inc process building footprints for implementation into geo-analytical services around emergency mapping and risk assessment of the population and infrastructure.
Semantic Analysis Models
The core Mapflow technology that helps cartographers to accelerate their work is semantic analysis.
đ Buildings Detection outlines the roofs of buildings in the high-res satellite images
Additional options:
- Classification by types of buildings â Typology is represented by the main classes: apartment buildings; private; industrial⊠(Reference for Urban Mapping classes)
- Building heights â building height estimation by the length of the shadow and the visible part of the wall. Shift to the building footprint
đ Extracting the Forest vegetation classes from RGB images of high resolution (2 meters) with classification by type, density and heights
đ Roads Detection â extracts the road mask from satellite images of high spatial resolution
đïž Construction Detection looks for construction sites by classification of tiles of high-res satellite images
Data sources
Unlike many platform projects, we didnât not start with Landsat and Sentinel open data. Many interesting applications can be done with this data, but itâs not suitable for detailed mapping due to the low spatial resolution. We are considering these satellites for new pipelines we are working on now, combined with high resolution imagery.
As Mapflow evolves, we plan to both integrate free sources and expand our premium commercial suite through partnerships with data providers.
Imagery available in Mapflow by default is Mapbox Satellite. The company announced the biggest global update by the acquisition of Maxarâs Vivid Basic.
By the way, processing by Mapbox Satellite can be used for mapping in Openstreetmap. You may be interested in learning more about the project âOpen Urban Mappingâ.
One of the premium sources provided in Mapflow is Maxar SecureWatch, which means access to global coverage and a whole bunch of satellite imagery basemap products. If you want to get advice on how to use Mapflow to work with premium sources or to upload your own data for processing, we will be happy to help you.
Getting started with Mapflow
The easiest way to get familiarized with the functionality of Mapflow â use the Web app.
After registration, each user receives 500 credits for free to get started.
The interface and services are described in documentation, but have an intuitive design. Also, after registration, each user can look at the âDemo processingâ powered by Mapflow.
We did our best to design all the necessary settings in a single dashboard. Watch how it looks like or try it yourself â Mapflow.ai
References
- Mapflow Platform
- Mapflow documentation
- Facebook page
- Linkedin page
- Telegram
- Github page