This repository contains a PyTorch implementation of the paper titled: Change Detection Based on Deep Siamese Convolutional Network for Optical Aerial Images (🔗 Yang Zhan, Kun Fu, et al.)
Two different views of an image are resized to a
Two feature tensors are output by running the forward pass twice through the CNN. A loss function based on a distance metric between these feature tensors is used to optimize the parameters of this architecture. The loss function is defined in the next section.
A distance metric
The loss function is defined as
where
where
SZTAKI AirChange Benchmark set: This Benchmark set contains 13 aerial image pairs of size 952x640 and resolution 1.5m/pixel and binary change masks (drawn by hand), which were used for evaluation in publications [1] and [2]. (🔗 Dataset)
[1] Cs. Benedek and T. Szirányi: ”Change Detection in Optical Aerial Images by a Multi-Layer Conditional Mixed Markov Model”, IEEE Transactions on Geoscience and Remote Sensing, vol. 47, no. 10, pp. 3416-3430, 2009
[2] Cs. Benedek and T. Szirányi: ”A Mixed Markov Model for Change Detection in Aerial Photos with Large Time Differences”, International Conference on Pattern Recognition (ICPR), Tampa, Florida, USA, December 8-11, 2008