Deep learning (DL) techniques for change detection have undergone rapid development in the last few years. However, it is still a challenge how to reduce massive network parameters and sufficiently fuse bi-temporal image features to improve detection accuracy. Therefore, this work proposes a novel and lightweight network based on feature interleaved fusion and bi-stage decoding (FFBDNet) for change detection. In the encoding stage, considering the application problems caused by a large number of network parameters, we use the more efficient EfficientNet as the backbone to extract the bi-temporal image features based on Siamese architecture. To fuse the bi-temporal image features and reduce interference from surrounding objects, we propose a feature interleaved fusion module (FIFM), which can interleave the shared feature information and the difference variance feature information. During the decoding stage, the fused features are split into two groups, and a novel bi-stage decoding framework is proposed to generate the accuracy change map gradually. Extensive experiments and ablation studies are validated on three public change detection datasets: the WHU-CD, LEVIR-CD, and SYSU-CD datasets. Compared to state-of-the-art (SOTA) methods, the experimental results demonstrate that the proposed FFBDNet produces a better balance between performance and model parameters. Specifically, the F1 values obtained for these three datasets are 93.27%, 91.11%, and 80.10%, respectively, and the model parameters of the network are just 2.85 M.

Change detection (CD) stands out as a pivotal yet challenging task in the interpretation of remote sensing images. Significant developments have been witnessed, particularly with the rapid advancements in deep learning techniques. Nevertheless, challenges such as incomplete detection targets and unsmooth boundaries remain as most CD methods suffer from ineffective feature fusion. Therefore, this paper presents a multi-scale gated fusion network (MSGFNet) to improve the accuracy of CD results. To effectively extract bi-temporal features, the EfficientNetB4 model based on a Siamese network is employed. Subsequently, we propose a multi-scale gated fusion module (MSGFM) that comprises a multi-scale progressive fusion (MSPF) unit and a gated weight adaptive fusion (GWAF) unit, aimed at fusing bi-temporal multi-scale features to maintain boundary details and detect completely changed targets. Finally, we use the simple yet efficient UNet structure to recover the feature maps and predict results. To demonstrate the effectiveness of the MSGFNet, the LEVIR-CD, WHU-CD, and SYSU-CD datasets were utilized, and the MSGFNet achieved F1 scores of 90.86%, 92.46%, and 80.39% on the three datasets, respectively. Furthermore, the low computational costs and small model size have validated the superior performance of the MSGFNet.

Change detection has become an attractive research topic in the field of remote sensing imagery in recent years. Despite significant advancements driven by deep learning techniques, most current methods predominantly rely on fully supervised strategies. These methods require the collection of a large number of pixel-level labels, which is quite time-consuming and label-intensive. To address that, we propose a weakly supervised change detection method, termed ESAM-CD, which leverages multi-scale Class Activation Map (CAM) fusion and a fine-tuned EfficientSAM’s image encoder. First, we construct a classification model employing image-level labels with a deep supervision strategy to generate high-quality multi-scale CAM. Subsequently, a multi-scale CAM fusion module is proposed to refine the boundaries of change targets by harnessing information from various scales. Then, we utilize EfficientSAM with powerful generalization capabilities as the backbone and fine-tune it using a low-rank adaptation (LoRA) strategy to establish a change detection network. In such a network, bi-temporal images and the generated pseudo-labels are fed into the network. Additionally, to overcome the reliance of EfficientSAM’s decoder on prompts, we propose a prompt-free decoder based on the general convolutional layers to predict change maps. Finally, we validate the effectiveness of the proposed ESAM-CD using two publicly available change detection datasets (i.e., WHU-CD and LEVIR-CD). Comprehensive experiments demonstrate that our method outperforms other weakly supervised change detection methods, achieving outstanding performance on both datasets.

Accurate image matching is a crucial prerequisite for multimodal data fusion in Earth observation and serves as a foundational technology for deep-space exploration. While Earth multimodal remote sensing images (EMRSIs) often suffer from severe nonlinear radiometric differences (NRD) and geometric deformations, planetary multimodal remote sensing images (PMRSIs) introduce additional challenges due to disparate sensing modalities and predominantly weak-texture surfaces. To address these cross-domain challenges, we propose a Modality-Invariant Structural Representation (MISR) framework designed for robust matching of both EMRSIs and PMRSIs. The MISR framework consists of two core modules. First, a novel image scale space is constructed using Weighted Side-window based Gradient Guided Filtering (WSGGF), which facilitates effective noise suppression while preserving essential structural and edge features. Second, a multi-order structural representation is introduced via the Multi-order Gradient Location and Orientation Histogram (MGLOH) descriptor, which integrates a weighted dominant orientation assignment with an enhanced spatial configuration. By fusing first- and second-order gradient responses, MGLOH offers improved rotation invariance and stronger feature discrimination. Extensive experiments conducted on both Earth and planetary datasets demonstrate that the proposed method achieves superior matching accuracy and robustness compared to state-of-the-art approaches.

Change detection (CD) is a fundamental task that is pivotal in understanding surface changes. Recently, change detection methods have advanced rapidly and attained impressive results, driven by deep learning technology. However, existing methods generally employ fusion modules with the same design for multi-level features, overlooking the inherent distinctions between low-level spatial features and deep-level semantic features generated by deep networks. To overcome this limitation, this paper proposes a novel change detection network, referred to as DACNet. This method introduces a divide-and-conquer fusion strategy designed to fuse multi-level features using different fusion strategies. Specifically, the widely used MobileNetV2 is employed within a dual-branch architecture to extract multi-level features from bi-temporal images. Subsequently, the proposed divide-and-conquer fusion strategy comprises two specialized modules: the change region localization module (CRLM) and the edge complementarity module (ECM), which are tailored to fuse deep-level semantic features and low-level spatial features, respectively. Additionally, to mitigate the unnecessary noise introduced by the conventional UNet architectures, attention gates are introduced into the UNet decoder to enhance the changed information and suppress background noises. Extensive experiments are conducted on three available change detection datasets: LEVIR-CD, Google-CD, and MSRS-CD. The proposed network achieved favorable results compared to the nine state-of-the-art methods across all experiments, improving the F1 score by 0.93%, 1.10%, and 0.81% on the LEVIR-CD, Google-CD, and MSRS-CD datasets, respectively. The code for this work will be available at https://github.com/wm-Githuber/DACNet.

Remarkable results have been achieved by DCNN based self-supervised depth estimation approaches. However, most of these approaches can only handle either day-time or night-time images, while their performance degrades for all-day images due to large domain shift and the variation of illumination between day and night images. To relieve these limitations, we propose a domain-separated network for self-supervised depth estimation of all-day images. Specifically, to relieve the negative influence of disturbing terms (illumination, etc.), we partition the information of day and night image pairs into two complementary sub-spaces: private and invariant domains, where the former contains the unique information (illumination, etc.) of day and night images and the latter contains essential shared information (texture, etc.). Meanwhile, to guarantee that the day and night images contain the same information, the domain-separated network takes the day-time images and corresponding night-time images (generated by GAN) as input, and the private and invariant feature extractors are learned by orthogonality and similarity loss, where the domain gap can be alleviated, thus better depth maps can be expected. Meanwhile, the reconstruction and photometric losses are utilized to estimate complementary information and depth maps effectively. Experimental results demonstrate that our approach achieves state-of-the-art depth estimation results for all-day images on the challenging Oxford RobotCar dataset, proving the superiority of our proposed approach.