CANet: A Combined Attention Network for Remote Sensing Image Change Detection

There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.

Abstract

Change detection (CD) is one of the essential tasks in remote sensing image processing and analysis. Remote sensing CD is a process of determining and evaluating changes in various surface objects over time. The impressive achievements of deep learning in image processing and computer vision provide an innovative concept for the task of CD. However, existing methods based on deep learning still have problems detecting small changed regions correctly and distinguishing the boundaries of the changed regions. To solve the above shortcomings and improve the efficiency of CD networks, inspired by the fact that an attention mechanism can refine features effectively, we propose an attention-based network for remote sensing CD, which has two important components: an asymmetric convolution block (ACB) and a combined attention mechanism. First, the proposed method extracts the features of bi-temporal images, which contain two parallel encoders with shared weights and structures. Then, the feature maps are fed into the combined attention module to reconstruct the change maps and obtain refined feature maps. The proposed CANet is evaluated on the two publicly available datasets for challenging remote sensing image CD. Extensive empirical results with four popular metrics show that the designed framework yields a robust CD detector with good generalization performance. In the CDD and LEVIR-CD datasets, the F1 values of the CANet are 3.3% and 1.3% higher than those of advanced CD methods, respectively. A quantitative analysis and qualitative comparison indicate that our method outperforms competitive baselines in terms of both effectiveness and robustness.

Related collections

Most cited references 24

Record: found
Abstract: found
Article: not found

Fully Convolutional Networks for Semantic Segmentation.

Trevor Darrell, Evan Shelhamer, Jonathan Long (2017)

Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, improve on the previous best result in semantic segmentation. Our key insight is to build "fully convolutional" networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet, the VGG net, and GoogLeNet) into fully convolutional networks and transfer their learned representations by fine-tuning to the segmentation task. We then define a skip architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional networks achieve improved segmentation of PASCAL VOC (30% relative improvement to 67.2% mean IU on 2012), NYUDv2, SIFT Flow, and PASCAL-Context, while inference takes one tenth of a second for a typical image.