Looking for Change? Roll the Dice and Demand Attention

Abstract

Change detection, i.e., the identification per pixel of changes for some classes of interest from a set of bi-temporal co-registered images, is a fundamental task in the field of remote sensing. It remains challenging due to unrelated forms of change that appear at different times in input images. Here, we propose a deep learning framework for the task of semantic change detection in very high-resolution aerial images. Our framework consists of a new loss function, a new attention module, new feature extraction building blocks, and a new backbone architecture that is tailored for the task of semantic change detection. Specifically, we define a new form of set similarity that is based on an iterative evaluation of a variant of the Dice coefficient. We use this similarity metric to define a new loss function as well as a new, memory efficient, spatial and channel convolution Attention layer: the FracTAL. We introduce two new efficient self-contained feature extraction convolution units: the CEECNet and FracTALResNet units. Further, we propose a new encoder/decoder scheme, a network macro-topology, that is tailored for the task of change detection. The key insight in our approach is to facilitate the use of relative attention between two convolution layers in order to fuse them. We validate our approach by showing excellent performance and achieving state-of-the-art scores (F1 and Intersection over Union-hereafter IoU) on two building change detection datasets, namely, the LEVIRCD (F1: 0.918, IoU: 0.848) and the WHU (F1: 0.938, IoU: 0.882) datasets.

Feature Decomposition-Optimization-Reorganization Network for Building Change Detection in Remote Sensing Images

“…The network learns a spectral-spatialtemporal feature representation to generate features with rich spatial-temporal information, and combines a recurrent neural network with CNN to obtain change information. Diakogiannis et al [38] propose two methods based on ResNet, i.e., self-attention fusion and relative attention fusion. The former is used to enhance the attention of the region of interest in a single image, while the latter focuses on the correlation between dual-temporal images.…”

Unlocking large-scale crop field delineation in smallholder farming systems with transfer learning and weak supervision

“…We briefly describe the FracTAL-ResUNet architecture and loss function here and refer the reader to Diakogiannis et al [35] and Waldner et al [31] for more details. A FracTAL-ResUNet has three main features, reflected in its name: (1) a self-attention layer called a FracTAL unit that is inserted into standard residual blocks, (2) skip-connections that combine the inputs and outputs of residual blocks (similar to in the canonical ResNet), and (3) an encoder-decoder architecture (similar to a U-Net).…”

“…1. We use an attention-based CNN [35] followed by watershed segmentation to delineate field instances accurately (F1-score = 0.93, MCC = 0.65, median IoU = 0.86) across India. Airbus imagery yields higher performance than PlanetScope imagery.…”

Semantic-aware Dense Representation Learning for Remote Sensing Image Change Detection

“…Transfer a pre-trained model allows leveraging the knowledge of a source dataset to enhance the performance of a downstream task. ImageNet pre-training is widely used in change detection [7][8][9][10][11][12][13] and shows superior to random initialization, especially in small data regimes. Considering the domain gap between ImageNet and RS images, a new trend is to pre-train on the remote sensing data to learn the in-domain representations [10, 14-17, 19, 50-53].…”

“…Despite the great success of contemporary deep learning (DL)-based CD methods [7], the lack of a large labeled CD dataset limits their generalization to real-world applications. One effective solution to handle the data insufficiency is to fine-tune the model from ImageNet pre-training [8][9][10][11][12][13].…”