Texture features based on the gray-level co-occurrence matrix (GLCM) can effectively improve classification accuracy in geographical analyses of optical remote sensing (RS) images, with the parameters of scale of the GLCM texture window greatly affecting the validity. By analyzing human visual attention characteristics for geo-texture cognition, it was found that there is a strong correlation between the texture scale parameters and the domain shape knowledge in a specified geo-scene. Therefore, a new approach for quickly determining the multi-scale parameters of the texture with the assistance of a geographic information system (GIS) and domain knowledge is proposed in this paper. First, the validity of domain knowledge from an existing GIS database is measured by spatial data mining algorithms, including spatial partitioning, image segmentation, and space-time system evaluation. Second, the general domain shape knowledge of each category is described by the GIS minimum enclosing rectangle indices and rectangular-degree indices. Then, the corresponding multi-scale texture windows can be quickly determined for each category by a correlation analysis with the shape indices. Finally, the Fisher function is used to evaluate the validity of the scale parameters. The experimental results show that the multi-scale value keeps a one-to-one relationship with the classified objects, and their value ranges are from a few to tens, instead of the smaller values of a traditional analysis; thus, effective texture features at such a scale can be built to identify categories in a geo-scene. In this way, the proposed method can increase the total number of categories for a certain geo-scene and reduce the classification uncertainty, as well as better meet the requirements of large-scale image geo-analysis. It also has as high a calculation efficiency and as good a performance as the traditional enumeration method.

Accurate road extraction from remote sensing images is of great importance, yet the complex backgrounds, significant scale differences, and detailed characteristics of road information pose a tremendous challenge to the semantic modeling capabilities of deep learning methods. Recent studies have shown impressive success with diffusion-based methods in semantic segmentation. However, when applied to road extraction tasks, the remarkable performance can be diminished by intricate terrains and unreasonable feature fusion. In this article, we reframe road extraction as a mask-generating task, progressively refining road mask predictions through a timestep-by-timestep denoising process. Specifically, we propose a novel adaptive frequency-aware latent diffusion framework named AFLDiff. The pretrained VQ-VAE encoder with coupled perceptual compression suppresses semantic features that are weakly associated with roads in remote sensing images, reconstructing the image into a low-dimensional latent space for training. The adaptive frequency-aware module filters out redundant noise from the noisy features and enhances useful high- and low-frequency components, bridging the gap between noisy and semantic features for better fusion. Extensive experimental results demonstrate that our method achieves state-of-the-art (SOTA) performance on the DeepGlobe, Massachusetts, and CHN6-CUG datasets, with cross-dataset validation further confirming its robustness and generalization capabilities.

Optical satellite images frequently suffer from thin clouds, degrading the data quality. A thin cloud correction method is developed based on complementary dark objects on multiple visible bands to address this problem. First, thin cloud images are divided into irregular sub-areas using the super pixel segmentation algorithm, enabling the proper identification of dark objects in spatial domain across different visible bands. A criterion is then established to classify the dark objects into two types, namely, absolute dark objects (ADOs) and relative dark objects (RDOs). Subsequently, the quantitative correlation of thin clouds between visible bands is estimated by adopting the ADOs. Dark objects present complementarity in visible bands; thus, the RDOs on one band are spatially densified by referencing the RDOs on the other visible bands. Thereby, a thin cloud map with fine spatial details is interpolated by using all the ADOs and RDOs on a band, and the correction procedure is performed through subtraction. Eight visible data captured by Landsat platforms are collected for simulated and real experiments to evaluate the method's performance. Three representative thin cloud correction approaches are selected for visual and quantitative comparisons. The proposed method can correct thin clouds effectively and restore various scenes accurately. The interpolated thin cloud maps show enhanced texture details and finer representation compared with the benchmarks. In addition, the advantages of dark object densification for thin cloud map generation and the limitations of the proposed method are investigated.