Which cnn layer for which change? a cnn adaptation approach for change detection in remote sensing data

The purpose of this paper is to experimentally study the adaptation of convolutional neural networks (CNN) to the problem of change detection in remote sensing data. Specifically, our goal is to explore the impact of each layer of CNN (low, medium, and high level) in capturing changes. To this end, two types of changes are studied, an artificial change and a real change. The results indicates that it is recommended to use specific CNN layers to detect specific changes. However, prior information on the size and characteristics of the changes are needed to make such a decision.