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-stream or dual-stream networks to extract change features, which is insufficient to extract and learn bitemporal change information thoroughly. This paper 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-stream 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 state-of-the-art CD performance on the above three datasets.

Dense prediction tasks, such as semantic segmentation, depth estimation, and surface normal prediction, can be easily formulated as per-pixel classification (discrete outputs) or regression (continuous outputs). This per-pixel prediction paradigm has remained popular due to the prevalence of fully convolutional networks. However, on the recent frontier of segmentation task, the community has been witnessing a shift of paradigm from per-pixel prediction to cluster-prediction with the emergence of transformer architectures, particularly the mask transformers, which directly predicts a label for a mask instead of a pixel. Despite this shift, methods based on the per-pixel prediction paradigm still dominate the benchmarks on the other dense prediction tasks that require continuous outputs, such as depth estimation and surface normal prediction. Motivated by the success of DORN and AdaBins in depth estimation, achieved by discretizing the continuous output space, we propose to generalize the cluster-prediction based method to general dense prediction tasks. This allows us to unify dense prediction tasks with the mask transformer framework. Remarkably, the resulting model PolyMaX demonstrates state-of-the-art performance on three benchmarks of NYUD-v2 dataset. We hope our simple yet effective design can inspire more research on exploiting mask transformers for more dense prediction tasks. Code and model will be made available ¹ .

Aquaculture has important economic and environmental benefits. With the development of remote sensing and deep learning technology, coastline aquaculture extraction has achieved rapid, automated, and high-precision production. However, some problems still exist in extracting large-scale aquaculture based on high-resolution remote sensing images: (1) the generalization of large-scale models caused by the diversity of remote sensing in breeding areas; (2) the confusion of breeding target identification caused by the complex background interference of land and sea; (3) the boundary of the breeding area is difficult to extract accurately. In this paper, we built a comprehensive sample database based on the spatial distribution of aquaculture, and expanded the sample database by using confusing land objects as negative samples. A multi-scale-fusion superpixel segmentation optimization module is designed to solve the problem of inaccurate boundaries, and a coastal aquaculture network is proposed. Based on the coastline aquaculture dataset that we labelled and produced ourselves, we extracted cage culture areas and raft culture areas near the coastline of mainland China based on high-resolution remote sensing images. The overall accuracy reached 94.64% and achieved a state-of-the-art performance.

Crop-type mapping is the foundation of grain security and digital agricultural management. Accuracy, efficiency and large-scale scene consistency are required to perform crop classification from remote sensing images. Many current remote-sensing crop extraction methods based on deep learning cannot account for adaptation effects in large-scale, complex scenes. Therefore, this study proposes a novel adaptive feature-fusion network for crop classification using single-temporal Sentinel-2 images. The selective patch module implemented in the network can adaptively integrate the features of different patch sizes to assess complex scenes better. TabNet was used simultaneously to extract spectral information from the center pixels of the patches. Multitask learning was used to supervise the extraction process to improve the weight of the spectral characteristics while mitigating the negative impact of a small sample size. In the network, superpixel optimization was applied to post-process the classification results to improve the crop edges. By conducting the crop classification of peanut, rice, and corn based on Sentinel-2 images in 2022 in Henan Province, China, the novel method proposed in this paper was more accurate, indicated by an F1 score of 96.53%, than other mainstream methods. This indicates our model’s potential for application in crop classification in large scenes.

Deep learning (DL)-based architectures have shown a strong capacity to identify changes. However, existing change detection (CD) networks still suffer from limited applicability when it comes to multi-scale targets and spatially misaligned objects. For the sake of tackling the above problems, a region-based feature fusion network (RFNet) for CD of very high spatial resolution (VHR) remote sensing images is proposed. RFNet uses a fully convolutional Siamese network backbone where a multi-stage feature interaction module (MFIM) is embedded in the dual encoder and a series of region-based feature fusion modules (RFFMs) is used to generate change information. The MFIM fuses features in different stages to enhance the interaction of multi-scale information and help the network better distinguish complex ground objects. The RFFM is built based on region similarity (RSIM), which measures the similarity of bitemporal features with neighborhoods. The RFFM can reduce the impact of spatially offset bitemporal targets and accurately identify changes in bitemporal images. We also design a deep supervise strategy by directly introducing RSIM into loss calculation and shortening the error propagation distance. We validate RFNet with two popular CD datasets: the SECOND dataset and the WHU dataset. The qualitative and quantitative comparison results demonstrate the high capacity and strong robustness of RFNet. We also conduct robustness experiments and the results demonstrate that RFNet can deal with spatially shifted bitemporal images.