Change detection (CD) is one of the essential tasks in Remote Sensing applications. Deep Learning (DL) methods for CD are typically categorized as either bitemporal or multitemporal, and methods focus on one kind of task. In this paper, we propose a dual-branch supervised CD method which uses time series of Remote Sensing optical data to simultaneously perform two tasks, identifying binary abrupt changes and multitemporal seasonal ones. The method relies on a 3D fully convolutional architecture and uses dilated convolutions as well as spatial and channel attention mechanisms to examine the spatial and temporal dimensions of the data. The method is tested on a time series of multispectral images acquired by Landsat-8. Results show that the proposed framework proves effective in detecting the two kinds of changes.

Deep learning (DL) approaches are widely used to improve change detection (CD). In many application domains, unsupervised DL CD methods are preferred since gathering multitemporal labeled samples is challenging. Many unsupervised CD methods use pre-trained DL models to extract multiscale features. This does not allow for preserving the spatial context information and the object structure in the multiscale hidden-layer features, thus obtaining poor performance in modeling multiresolution changes. In this article, we propose two hierarchical loss functions to train multiscale hidden-layer features and preserve their spatial context information. The multiscale hidden-layer feature maps extracted from the model are used in an unsupervised multiscale CD method. We present two possible hierarchical loss functions. The first one considers all the model layers during the training by comparing the mirrored couples of encoder–decoder hidden-layer features, while the second one aims to preserve the geometrical details using a multiresolution-based loss function. After training, the CD method uses a feature selection (FS) based on structure-similarity-index (SSIM) to keep only the most informative hidden-layer feature maps. We tested the proposed method on bi-temporal multispectral images acquired by Landsat-8 representing a burned area and Sentinel-2 images representing a deforested area.

In the literature, Change Detection (CD) methods use the information from heterogeneous sensors to detect the changes more effectively with respect to single-sensor methods. Among them, Deep Learning (DL) CD methods try to learn a common feature domain or perform a domain adaptation, but many still present domain gaps between multi-sensor feature maps. We propose a DL CD method that learns multiscale feature maps in a domain common to multi-modal SAR images using a Cycle Generative Adversarial Network (CycleGAN). We apply a code-alignment loss function to reduce the domain gap between the feature maps derived from the multi-sensor SAR images. The multi-scale feature maps of the CycleGAN generators are processed to derive the change map. Preliminary experiments performed by processing bi-temporal SAR images acquired from COSMO-SkyMed and SAOCOM showed promising results.

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 multitemporal case. A possible alternative is leveraging methods that exploit transfer learning for CD by reusing DL models pretrained 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 multiresolution 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 bitemporal images to obtain and compare multiresolution bitemporal feature maps. These feature maps are then analyzed by a feature-selection technique to select the most discriminant ones. Furthermore, an aggregated multiresolution difference image is computed and used for a detail-preserving multiscale CD. In the context of this CD approach, we propose two alternative strategies to retrieve multiscale reliability maps. We tested the proposed method on bitemporal 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.

Deep Learning (DL) methods are widely used for Change Detection (CD) in multi-temporal Remote Sensing (RS) images. The recently reported unsupervised DL CD methods alleviate the problem of the labeled data collection affecting the supervised ones. Many of them exploit the DL models (e.g., Convolutional Autoencoder (CAE)) as a feature extractor and use the retrieved features to detect the changes. However, these features do not efficiently preserve the geometrical details, and they do not optimize the selection of informative features for change detection. We propose an unsupervised DL CD method that exploits the features extracted by a CAE trained with a super-resolution based loss function. The loss function allows the CAE to be trained to reconstruct the spatial information thus generating features preserving the geometrical details. The proposed method exploits a feature selection based on the Structured Similarity Index (SSIM) to perform a texture analysis and chooses couples of bi-temporal features providing relevant information about changes. We tested the proposed method on a couple of bi-temporal Landsat-8 images representing a burned area near Granada, Spain.