Unsupervised Change Detection Using Convolutional-Autoencoder Multi-resolution Features

The use of Deep Learning (DL) methods for Change Detection (CD) is currently dominated by supervised models that require a large number of labeled samples. However, these samples are difficult to acquire in the multi-temporal case. A possible alternative is leveraging methods that exploit transfer learning for CD by reusing DL models pre-trained for other tasks. However, the performance of the transfer-learning-based models decreases as much as the target images differ from the ones used for training the model. To overcome this limit, we propose an unsupervised CD method that exploits multi-resolution deep feature maps derived by a Convolutional Autoencoder (CAE). It automatically learns spatial features from the input during the training phase without requiring any labeled data. The proposed method processes the bi-temporal images to obtain and compare multi-resolution bi-temporal feature maps. These feature maps are then analyzed by a feature-selection technique to select the most discriminant ones. Furthermore, an aggregated multi-resolution difference image is computed and used for a detail-preserving multi-scale change detection. In the context of this CD approach, we propose two alternative strategies to retrieve multi-scale reliability maps. We tested the proposed method on bi-temporal multispectral images acquired by Landsat-5 and Landsat-8 representing burned areas and Sentinel-2 images representing deforested areas. Results confirm the effectiveness of the proposed CD technique.