Object Identification on satellite images using Neural Networks with Tensorflow. Datasets used: COWC, SpaceNet Buildings Dataset. SIMRDWN framework was used for our experiments. Technologies used: Python, Keras (Deep Learning python library), and Tensorflow. Objective: The model is trained upon annotated images containing cars and building footprints, giving the ability to identify their boundaries.
YOLT is an extension of the YOLO v2 framework that can evaluate satellite images of arbitrary size, and runs at ~50 frames per second. Current applications include vechicle detection (cars, airplanes, boats), building detection, and airport detection.
- Install SIMRDWN v1 as per your system requirements. Follow the installation instructions on the link.
- Download COWC dataset. Use ground truth sets as input images.
- Download SpaceNet building footprints dataset. Needs a AWS account. Suggested client browser: CloudBerry Explorer for Amazon S3
- Follow script_cmds.sh commands for reproducing our experiments. In results folder you will find all the logs of our trains and tests.
simrdwn_v1-help.txt and simrdwn_v2-help.txt contains a complete list of parameters used by SIMRDWN framework.
Acknowledgments
This project was developed during the Spring Semester of 2019 for the module M111 - Big Data.
Authors
Maria-Evangelia Pavlopoulou
Georgios Kalampokis