Answers are grounded in this paper and cite their sources.
Deserts are the driest ecosystems on Earth, with land desertification posing significant challenges to the sustainable development of inland ecological environments and human activities. However, land cover in desert regions is complex, yet sparsely distributed, and there is a lack of publicly available datasets to study desert reclamation. On the other hand, deep learning-based semantic change detection (SCD) algorithms are of critical importance for monitoring and evaluating desert reclamation. Consequently, we create a desert semantic change detection (DSCD) dataset, which contains 10 000 pairs of high-resolution remote sensing images from the northwest region of the Three-North Shelterbelt Forest, covering four distinct land cover change classes. We then propose a cross-dimensional information enhancement network (CIENet) to achieve cross-dimensional fusion of high-level semantic information and low-level spatial information for high-precision DSCD. The CIENet is capable of employing any backbone for the extraction of remote sensing features. In CIENet, Cross Dimensional Feature Fusion (CDFF) is designed to utilize high-level information to strengthen the semantic features of low-level spatial information, thereby obtaining multidimensional semantic features. Bitemporal difference information aggregation is designed to integrate independent bitemporal features to obtain a different feature. Parallel semantic information enhancement is designed to reinforce semantic features by integrating both global and local information, thereby obtaining fine-grained semantic features. Experimental results demonstrate that the proposed CIENet surpasses SOTA methods in accuracy, detail preservation, and robustness, providing a scientific foundation to assess and guide SCD.
No takes yet. Share an insight, caveat, or question.
Huang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: