https://github.com/vbhavank/Siamese-neural-network-for-change-detection

This repository contains the python code for a Siamese neural network to detect changes in aerial images using Tensorflow.
https://github.com/vbhavank/Siamese-neural-network-for-change-detection

Last synced: 12 days ago
JSON representation

This repository contains the python code for a Siamese neural network to detect changes in aerial images using Tensorflow.

• Host: GitHub
• URL: https://github.com/vbhavank/Siamese-neural-network-for-change-detection
• Owner: vbhavank
• License: gpl-3.0
• Created: 2018-05-10T18:14:34.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2019-11-27T14:56:27.000Z (almost 5 years ago)
• Last Synced: 2024-02-17T14:36:14.176Z (9 months ago)
• Language: Python
• Size: 187 KB
• Stars: 81
• Watchers: 6
• Forks: 37
• Open Issues: 6
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        

   

Implementation of "SIAMESE NETWORK WITH MULTI-LEVEL FEATURES FOR PATCH-BASED CHANGE DETECTION IN SATELLITE IMAGERY"

[1] Faiz Ur Rahman, Bhavan Kumar Vasu, Jared Van Cor, John Kerekes, Andreas Savakis, "Siamese Network with Multi-level Features for Patch-based Change Detection in Satellite Imagery", IEEE SigPort, 2018. [Online]. Available:https://sigport.org/documents/siamese-network-multi-level-features-patch-based-change-detection-satellite-imagery. Accessed: Feb. 21, 2019.

We present a patch-based algorithm for detecting structural changes in satellite imagery using a Siamese neural network. The two channels of our Siamese network are based on the VGG16 architecture with shared weights. Changes between the target and reference images are detected with a fully connected decision network that was trained on DIRSIG simulated samples and achieved a high detection rate. Alternatively, a change detection approach based on Euclidean distance between deep convolutional features achieved very good results with minimal supervision.

Dependencies required

   1)Tensorflow

   2)Keras with tensorflow background

   3)Numpy

   4)Keras.utils

   5)numpy_utils

   6)Python 2.7

Data

   Few sample data in is present in image pairs

   Unzip the file

   Names starting with AChip has a corresponding ANeg these are the the pairs

   for example

      AChip1,ANeg1 becomes a pair

      AChip2.ANeg2 becomes a pair

Testing 

   Siamese_predict.py is used for testing

   open command line and type

   python Siamese_predict.py

   It will ask for 1st image chip choose the image pairs as described above 

   Do the same for 2nd image chip

Output will be in command line Change or No change

Please Cite our work using the bib below.

@inproceedings{rahman2018siamese,

  title={Siamese Network with Multi-Level Features for Patch-based Change Detection in Satellite Imagery},

  author={Rahman, Faiz and Vasu, Bhavan and Van Cor, Jared and Kerekes, John and Savakis, Andreas},

  booktitle={2018 IEEE Global Conference on Signal and Information Processing (GlobalSIP)},

  pages={958--962},

  year={2018},

  organization={IEEE}

}