Abstract
Abstract. Change detection applications from satellite imagery can be a very useful tool in monitoring human activities and understanding their interaction with the physical environment. In the past few years most of the recent research approaches to automatic change detection have been based on the application of Deep Learning techniques and especially on variations of Convolutional Neural Network architectures due to their great representational capacity and their state-of-the-art performance in visual tasks such as image classification and semantic segmentation. In this work we train and evaluate two CNN architectures, UNet and UNet++, on a change detection task using Very High-Resolution satellite images collected at two different time epochs. We also examine and analyse the effect of two different loss functions, a combination of the Binary Cross Entropy Loss with the Dice Loss, and the Lovász Hinge loss, both of which were specifically designed for semantic segmentation applications. Finally, we experiment with the use of data augmentation as well as deep supervision techniques to evaluate and quantify their contribution in the final classification performance of the different network architectures.
Highlights
The application of a reliable Change Detection (CD) framework can be an invaluable tool in understanding the relationships and interactions between human activities and the physical environment, as well as for map updating and urban monitoring
In this work we aim to evaluate the use of both UNet and UNet++ architectures for change detection on satellite images and to compare the results produced by training an original UNet architecture on a Change Detection dataset with the results achieved by the UNet++ architecture
We build on the work of Peng et al (2019) with the goal to evaluate the effect of different architectural choices (UNet and UNet++ encoder-decoder architectures) in combination with different loss functions on the performance of the trained networks for change detection applications
Summary
The application of a reliable Change Detection (CD) framework can be an invaluable tool in understanding the relationships and interactions between human activities and the physical environment, as well as for map updating and urban monitoring. The latest and most successful approaches on automatic change detection from satellite images relate mainly to using Deep Convolutional Neural Networks (DCNN) with an encoder-decoder architecture that directly produce a ‘Change’‘No Change’ label for each pixel of the original image. Such an architecture was first introduced by Fully Convolutional Networks (FCNs) (Long et al, 2015) and since many successful variants have been proposed including SegNet (Badrinarayanan et al, 2017), UNet (Ronneberger et al, 2015) and UNet++ (Zhou et al, 2018). We evaluate the effectiveness of training using different loss functions as well as the benefits of data augmentation
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
