Object-Based Change Detection in Satellite Images Combined with Neural Network Autoencoder Feature Extraction

Change detection is a challenging task in the field of remote sensing. While most applications use supervised change detection and classification techniques, it still remains a difficult task to create a training database for generic applications. In this paper, we propose an end-to-end unsupervised change detection and clustering baseline. The presented baseline firstly deploys a neural network autoencoder for feature extraction and comparison to detect some meaningful changes. These changes are further segmented with 3D graph-based techniques and clustered. The presented algorithm gives promising results and is fully unsupervised.