Drones 🚁 ❤️ RoboSat 🤖

[maning]

Drones 🚁 ❤️ RoboSat 🤖

RoboSat

An end-to-end pipeline written in Python 3 for feature extraction from aerial and satellite imagery. Features can be anything visually distinguishable in the imagery.

At its core RoboSat is using state of the art fully convolutional neural network architectures for semantic segmentation.

Drones at OpenAerialMap

Data prep

Positive and negative training data

Diversifying training data sources

Arteche, Eastern Samar	Bacong River, Culasi, Antique	Narvacan Sulvec Port Road	Sugarcane LAREC Mill District, Pampanga
2014-12-01 4 cm Unknown sensor	2017-05-24 3 cm DJI Mavic Pro	2018-03-14 4 cm DJI Phantom 4	2017-03-30 5 cm EB Sensefly

Training the model

image	prediction by epoch

Running the prediction

https://maning.github.io/robosat-viz/can-avid.html

Good detection. Most buildings were detected by the model. Adjacent buildings with no visible separation tend to be detected as one contiguous shape.

Bad detection of shape. Damaged buildings due to typhoon were detected but the shape is not very well defined.

See also

• https://www.openstreetmap.org/user/maning/diary/44462
• https://github.com/maning/robosat-viz
• https://github.com/mapbox/robosat

Meta

• A lightning talk presented to Pista ng Mapa 2019