Change detection (CD) is the process of identifying changes in the category or attributes of ground objects by observing remote sensing images (RSI) taken at different times. In recent years, transformers have shown great potential in CD. However, current transformer-based CD networks have not fully exploited the capabilities of the transformer, especially when fusing features from bi-temporal and multi-stages. In this work, a pure transformer-based CD network (TFN) is built to fuse features better. Specifically, we build a Siamese CD network based on the transformer. Bi-temporal features are fused through a Shifted Window Fusion Model (SWFM) to address the misalignment between the features. In the decoding phase, a Multi-Scale Transformer Decoder (MSTD) is introduced to generate more complete change masks. The proposed method is validated on the WHU and SECOND datasets, demonstrating state-of-the-art performance (SOTA).

To overcome the effect of speckle noise on SAR image change detection, a superpixel segmentation algorithm based on speckle reducing anisotropic diffusion model is proposed and applied for object-oriented SAR image change detection. Based on the traditional modeling methods of the non-similarity between pixels and seed points by combining spectral distance and spatial distance in superpixel segmentation, the concept of diffusion flux is proposed to simulate the continuous and bounded evolution of the membership of pixels and seed points in the image plane lattice. Considering the effect of speckle noise and the demand of segmenting different shape surface features in complex scenes, the speckle reducing anisotropic diffusion is used to model the diffusion flux. After superpixel segmentation of dual-temporal remote sensing images, an overlay technology is adopted to obtain the finer results. Finally, the change detection result is generated based on the superpixelized difference image by the classical fuzzy clustering algorithm FCM. The experiments carried out on Sentinel-1 SAR images by comparing algorithms fully demonstrate the effectiveness of the proposed algorithm.

This study proposed a new application paradigm of time series SAR data for sugarcane mapping. First, the LOESS smoothing technique was exploited to reconstruct time series SAR data and reduce SAR noise in the time domain. Second, temporal importance was evaluated using RF MDA ranking, and basic parcel units were obtained only based on multitemporal SAR images with high importance values. At last, the parcel-based classification method, combining time series smoothing SAR data, RF classifier, and basic parcel units, was used to generate a sugarcane extent map without unreasonable sugarcane spots. The proposed paradigm was applied to map sugarcane cultivation in Suixi County, China. Results showed that the proposed paradigm was able to produce an accurate classification map with an overall accuracy of 96.09% and a Kappa coefficient of 0.91. Compared with the pixel-based classification result with original time series SAR data, the new paradigm performed much better in reducing the "salt and pepper" spots and improving the completeness of the sugarcane plots. Especially, the unreasonable non-vegetation spots in the sugarcane map were completely eliminated. The results demonstrated the efficacy of the new paradigm for mapping sugarcane cultivation.

Cloud detection is a crucial step in the analysis and processing of optical remote sensing satellite imagery. Existing methods often have large parameter sizes, high computational complexity, and experience a rapid drop in model accuracy when transferred to different sensors. We propose a cloud detection method based on Efficient Transformer and Neural Style Transfer, a lightweight cloud detection network that can be used across different sensors without requiring additional labeled data. Our contribution focus on two main aspects: 1.We design a lightweight cloud detection model that reduces redundant parameters without compromising model accuracy; 2.We introduce a Neural Style Transfer module for cross-sensor cloud detection, aiming to align cloud features from different sensors with training images. Experiments on the 38-Cloud dataset demonstrate that our proposed method achieves state-of-the-art performance while maintaining a small parameter size and floating-point computation. In cross-sensor cloud detection experiments, the inclusion of the Neural Style Transfer module significantly enhances the model’s capability for cross-sensor cloud detection.

Change detection (CD) based on high-resolution remote sensing images can be used to monitor land cover changes, which is an important and challenging topic in the remote sensing field. In recent years, with the development of deep learning, CD methods based on deep learning have achieved good results in the field of CD. However, most current CD methods use single- or dual-stream networks to extract change features, which is insufficient to extract and learn bitemporal change information thoroughly. This article proposes a triple-stream network (TSNet) with cross-stage feature fusion for CD in high-resolution bitemporal remote sensing images. First, to obtain highly representative deep features in the original image, we perform feature extraction on bitemporal remote sensing images and their concatenated image with a dual-stream encoder and a single-stream encoder, respectively. Then, the bitemporal multiscale features extracted by the dual-stream encoder are input into a multistage bidirectional convolutional gated recurrent unit (MSBC_GRU) feature fusion module, allowing the network to learn the change information in a cross-stage manner. In addition, we use a dual-channel attention module to fuse the features extracted by dual- and single-stream encoders, improving the network’s ability to discriminate changed features. The effectiveness of TSNet is demonstrated with three publicly available CD datasets. The extensive experimental results demonstrate that the proposed method achieves the state-of-the-art CD performance on the above three datasets.

The land-use map is an important data that can reflect the use and transformation of human land, and can provide valuable reference for land-use planning. For the traditional image classification method, producing a high spatial resolution (HSR), land-use map in large-scale is a big project that requires a lot of human labor, time, and financial expenditure. The rise of the deep learning technique provides a new solution to the problems above. This paper proposes a fast and precise method that can achieve large-scale land-use classification based on deep convolutional neural network (DCNN). In this paper, we optimize the data tiling method and the structure of DCNN for the multi-channel data and the splicing edge effect, which are unique to remote sensing deep learning, and improve the accuracy of land-use classification. We apply our improved methods in the Guangdong Province of China using GF-1 images, and achieve the land-use classification accuracy of 81.52%. It takes only 13 hours to complete the work, which will take several months for human labor.