August, 2023
Authors: Candela Sol Pelliza & Rodrigo Brust Santos
Application Development (Object-Based Image Analysis)
- Dr. Prof. Dirk Tiede & Dr. Prof Martin Sudmanns
This repository is the development of the final project for Application Development lecture.
In the Remote Sensing industry, OBIA is a technique that aims to utilize objects instead of pixels when analyzing an image.
From a satellite scene, one must segment pixel values in groups with similar values, following up for the classification.
It is a powerful approach, since it is fiasible to utilize several bands and rasters, such as the regular RGB but also DEM and DSM, leading to more reliable classification - and avoiding salt-peper effect.
In the EO*GI industry, there are a lot of softwares and resources that are very convinient, however an expensive subscription is necessary. Also, finding a good OBIA workflow is an exhausthing process.
Having that in mind, this project has two main objectives:
1 - Apply the concepts of OBIA with Python on a 5-band-scene, using R, G, B, and NIR bands in addition to a DSM.
-
Classify high/low vegetation, road and houses based on NDVI and height.
-
Export classified segments to geojson.
2 - Provide a resource for students and industry players from an open-source software.
To achieve the result, there were a couple of steps:
1 - Segment the image with scikit-image quickshift algorithm.
2 - Create the mean of rgb for each object
3 - Calculate the mean NDVI and Height for each object
4 - Create rules and assign classes
- In total, there were 5 classes:
unclassified
,low vegetation
,trees
,roads
andbuildings
.
5 - Generate the classified image