Deep learning models utilizing spatial features have demonstrated outstanding performance in remote sensing change detection. However, current deep learning methods relying on spatial features still exhibit limitations in accurately detecting changed edge regions and insignificant areas. To overcome this limitation, we propose the multiscale spatial-frequency feature fusion network (MSFNet) for change detection. The network aims to enhance the detection capability of change edge and insignificant regions. In the encoding stage, we introduce a frequency difference feature extraction module to enhance the frequency difference features by extracting high-frequency information that represents changes and low-frequency information that represents invariance at each feature scale. This approach effectively mitigates the issue of incomplete change information caused by relying solely on a single spatial feature dimension for feature extraction. In the decoding stage, a spatial frequency difference fusion module is introduced to map the extracted frequency difference information back to the spatial domain. This module enhances the network's semantic modeling capability for change regions while mitigating the degradation of edge detection accuracy resulting from information loss during feature extraction. The experimental results demonstrate that MSFNet has remarkable performance on both the LEVIR-CD and WHU-CD datasets, with an F1 score of 91.66 and an IoU of 84.41 on the LEVIR-CD dataset, and an F1 score of 90.87 and an IoU of 82.80 on the WHU-CD dataset.

Most heterogeneous change detection methods based on transfer learning may not yield satisfactory results due to the lack of comprehensive utilization of the global and local characteristics of the image. In this paper, we propose an unsupervised heterogeneous change detection method based on iterative joint global-local translation (IJGLT). The two heterogeneous images are firstly segmented into superpixels with the same boundary. Unlike the transfer learning methods for classification tasks that directly reduce the distance of the distribution of all data, we construct a translation function to decrease the difference between unchanged superpixels while increasing the difference of changed superpixels. During the translation process, the overall distribution is constrained using pseudo change labels, and the local manifold preserving is utilized to maintain the spatial structure and local adjacency relationship of superpixels. By comparing the translated superpixels of the heterogeneous images, the change map is obtained and then refined with pseudo change labels taking an iterative approach until convergence. Using the public and self-made optical-SAR heterogeneous datasets with different resolutions, the experimental results demonstrate that IJGLT is superior to several typical comparison methods with a maximum overall accuracy of 0.98 and a kappa coefficient of 0.9.

Compared to two-view reconstruction, multi-view imagery leverages redundant information to mitigate the effects of occlusion and noise. Deep learning-based multi-view stereo (MVS) methods are primarily applicable to tri-stereo data captured simultaneously and rely heavily on training samples. Traditional MVS methods typically rely on simple filtering and weighting techniques for DSM fusion based on image pair selection and pair-wise stereo matching, which are usually affected by poor image pairs and fail to fully exploit the complementary advantages of DSMs. To address these limitations, this paper proposes a novel DSM fusion method incorporating learning-based dual-scale enhanced confidence for 3D reconstruction from multi-view satellite imagery. First, a generalized stereo matching method is adopted, which considers radiometric differences and small feature variations. Next, auxiliary information generated during pair-wise stereo reconstruction is utilized to construct a high-dimensional confidence vector that includes classical confidence measures and a newly designed topological structure relationship consistency (TSRC) measure. Then, a guided regularized random forest regressor is employed to identify influential confidence measures and establish their correlation with reconstruction accuracy, leading to the estimation of enhanced confidence. Additionally, to preserve fine details and boundary information, the dual-scale enhanced confidence is introduced to facilitate cross-scale DSM fusion. Finally, each view is sequentially treated as the master view to obtain DSMs, which are then fused to produce the final DSM. Experimental results demonstrate that the proposed method achieves superior performance across various datasets, including tri-stereo Beijing-3 data acquired nearly simultaneously, multi-view WorldView-3 data captured at different times, and self-made tri-stereo data collected from different sensors at different times. The proposed method achieves an average MAE of 1.14 m, RMSE of 2.16 m, MHE of 0.47 m, and COMP of 75.13%, outperforming several mainstream methods.

Addressing the challenge of multi-class change detection with a small sample size, a change vector analysis with adaptive weight Gaussian mixture model (CVA-AWGMM) is proposed in this article. Initially, in order to avoid errors caused by geometric distortion and imprecise alignment, we employed a neighborhood search approach when calculating the different map. Instead of direct pixel-to-pixel comparisons, we compensated corresponding pixels in the images before and after the change by finding the minimum difference within a specified pixel range. Following this, we employ the fundamental framework of CVA to extract magnitude and angle features from the change vectors, facilitating the identification of both changed and unchanged regions through threshold segmentation. Based on the above binary change detection result, a semi-supervised Gaussian mixture model is used for further change class differentiation. Recognizing the inherent challenges of effectively training classifier model with a small sample size, the clustering features in a large number of unlabeled samples and the supervised information from a few labeled samples are simultaneously utilized to co-construct the objective function of the model. Meanwhile, considering the complementary properties of magnitude and angle features, the labeled samples are used to adaptively weight the features of both to further improve the accuracy of the method. Experiments were conducted on two public data sets and one self-made data set, and the results demonstrate that the proposed CVA-AWGMM outperforms several typical methods.

In response to the existing challenges in semantic change detection for remote sensing images, such as weak spatiotemporal correlation and insufficient utilization of local neighborhood information, this paper proposes a semantic change detection network based on hierarchical local-sparse attention (HLSNet). The network combines a fully convolutional network (FCN) with a deep transformer structure to leverage the advantages of local feature extraction and long-range information connection. Next, a hierarchical local-sparse attention is proposed to exploit the neighborhood characteristics of target pixels using a dual-window attention mechanism, the aim is to increase the receptive field while minimizing the interference of redundant information. By focusing on all tokens within a smaller window and dynamically selecting key tokens within a larger window for attention calculation, this two-tiered attention approach allows the model to handle details while capturing broader contextual information. The small window provides tightly related local information, while the larger window offers relevant but potentially more distant information, achieving a hierarchical processing of information from local to long-range. In order to facilitate more comprehensive interaction between the features of pre- and post-change images, each transformer block in the network employs a strategy of concatenating self-attention and cross-attention. This approach better captures the spatiotemporal correlations and feature integration, thus achieving efficient and precise change detection. HLSNet achieves the highest accuracy on the two commonly used SCD datasets, SECOND and Landsat-SCD, with F _scd values reaching 62.53% and 91.67%, respectively.

The water cloud model (WCM) has been widely used to retrieve vegetation parameters, such as biomass and soil moisture. To address difficulties in the measurement of soil moisture data and inaccurate estimation of soil scattering in the original model, a semiempirical calibration of the WCM for maize biomass retrieval is proposed in this letter. In it, a change detection approach is used to estimate soil moisture and the effect of surface roughness, calculated from the polarization difference, is added to the soil layer scattering. By combining the two polarization retrieval results and setting the weight coefficient of the VV polarization to a value greater than that of the VH polarization, the optimal retrieval results for maize biomass are obtained for the study area. The accuracy of the calibration of the WCM is verified using the backscatter coefficients from Sentinel-1 data and ground-based maize biomass measurements. It is found that the root-mean-square error (RMSE) and the coefficient of determination (R²) are 1.642 kg/m² and 0.803, respectively, between the calibration of the WCM results and the measurements. These results demonstrate the application potential of the C-band synthetic aperture radar data using the semiempirical calibration of the WCM for retrieving maize biomass over large-scale areas.