Search Results (2)

Land-use and land-cover (LULC) change detection (CD) is a pivotal research area in remote sensing applications, posing a significant challenge due to variations in illumination, radiation, and image noise between bi-temporal images. Currently, deep learning solutions, particularly convolutional neural networks (CNNs), represent the state of the art (SOTA) for CD. However, CNN-based models require substantial amounts of annotated data, which can be both expensive and time-consuming. Conversely, acquiring a large volume of unannotated images is relatively easy. Recently, self-supervised contrastive learning has emerged as a promising method for learning from unannotated images, thereby reducing the need for annotation. However, most existing methods employ random values or ImageNet pre-trained models to initialize their encoders and lack prior knowledge tailored to the demands of CD tasks, thus constraining the performance of CD models. To address these challenges, we introduce a novel feature-differencing-based framework called Barlow Twins for self-supervised pre-training and fine-tuning in CD (BTCD). The proposed approach employs absolute feature differences to directly learn unique representations associated with regions that have changed from unlabeled bi-temporal remote sensing images in a self-supervised manner. Moreover, we introduce invariant prediction loss and change consistency regularization loss to enhance image alignment between bi-temporal images in both the decision and feature space during network training, thereby mitigating the impact of variation in radiation conditions, noise, and imaging viewpoints. We select the improved UNet++ model for fine-tuning self-supervised pre-training models and conduct experiments using two publicly available LULC CD datasets. The experimental results demonstrate that our proposed approach outperforms existing SOTA methods in terms of competitive quantitative and qualitative performance metrics. Full article

Self-supervised learning (SSL) has emerged as a promising paradigm for learning flexible speech representations from unlabeled data. By designing pretext tasks that exploit statistical regularities, SSL models can capture useful representations that are transferable to downstream tasks. Barlow Twins (BTs) is an SSL technique inspired by theories of redundancy reduction in human perception. In downstream tasks, BTs representations accelerate learning and transfer this learning across applications. This study applies BTs to speech data and evaluates the obtained representations on several downstream tasks, showing the applicability of the approach. However, limitations exist in disentangling key explanatory factors, with redundancy reduction and invariance alone being insufficient for factorization of learned latents into modular, compact, and informative codes. Our ablation study isolated gains from invariance constraints, but the gains were context-dependent. Overall, this work substantiates the potential of Barlow Twins for sample-efficient speech encoding. However, challenges remain in achieving fully hierarchical representations. The analysis methodology and insights presented in this paper pave a path for extensions incorporating further inductive priors and perceptual principles to further enhance the BTs self-supervision framework. Full article