2 February 2024 in IEEE Transactions on Geoscience and Remote Sensing
Remote sensing (RS) images change detection (CD) plays a crucial role in monitoring surface dynamic. However, current deep learning (DL)-based CD methods still suffer from pseudo changes and scale variations due to inadequate exploration of temporal differences and under-utilization of multiscale features. Based on the aforementioned considerations, a spatiotemporal enhancement and interlevel fusion network (SEIFNet) is proposed to improve the ability of feature representation for changing objects. Firstly, the multilevel feature maps are acquired from Siamese hierarchical backbone. To highlight the disparity in the same location at different times, the spatiotemporal difference enhancement modules (ST-DEM) are introduced to capture global and local information from bitemporal feature maps at each level. Coordinate attention and cascaded convolutions are adopted in subtraction and connection branches, respectively. Then, an adaptive context fusion module (ACFM) is designed to integrate interlevel features under the guidance of different semantic information, constituting a progressive decoder. Additionally, a plain refinement module and a concise summation-based prediction head are employed to enhance the boundary details and internal integrity of CD results. The experimental results validate the superiority of our lightweight network over 8 state-of-the-art (SOTA) methods on LEVIR-CD, SYSU-CD and WHU-CD datasets, both in accuracy and efficiency. Also, the effects of different types of backbones and differential enhancement modules are discussed in the ablation experiments in details. The code will be available at https://github.com/lixinghua5540/SEIFNet.
17 November 2023 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Optical and synthetic aperture radar (SAR) images provide complementary information to each other. However, the heterogeneity of same-ground objects brings a large difficulty to change detection (CD). Correspondingly, transformation-based methods are developed with two independent tasks of image translation and CD. Most methods only utilize deep learning for image translation, and the simple cluster and threshold segmentation leads to poor CD results. Recently, a deep translation-based CD network (DTCDN) was proposed to apply deep learning for image translation and CD to improve the results. However, DTCDN requires the sequential training of the two independent subnetwork structures with a high computational cost. Toward this end, a concatenated deep-learning framework, multitask change detection network (MTCDN), of optical and SAR images is proposed by integrating the CD network into a complete generative adversarial network. This framework contains two generators and discriminators for optical and SAR image domains. Multitask refers to the combination of image identification by discriminators and CD based on an improved UNet++. The generators are responsible for image translation to unify the two images into the same feature domain. In the training and prediction stages, an end-to-end framework is realized to reduce cost. The experimental results on four optical and SAR datasets prove the effectiveness and robustness of the proposed framework over eight baselines.