Search

Effective extraction of building from remote sensing images requires both global and local information. Despite convolutional neural networks (CNNs) excelling at capturing local details, their intrinsic focus on local operations poses challenge in effectively extracting global features, especially in the context of large‐scale buildings. In contrast, transformers excel at capturing global information, but compared to CNNs, they tend to overly rely on large‐scale datasets and pre‐trained parameters. To tackle the challenge, this paper presents the multi‐scale and dual‐branch feature fusion network (MDTrans). Specifically, the CNN and transformer branches are integrated in a dual‐branch parallel manner during both encoding and decoding stages, local information for small‐scale buildings is extracted by utilizing Dense Connection Blocks in the CNN branch, while crucial global information for large‐scale buildings is effectively captured through Swin Transformer Block in the transformer branch. Additionally, Dual Branch Information Fusion Block is designed to fuse local and global features from the two branches. Furthermore, Multi‐Convolutional Block is designed to further enhance the feature extraction capability of buildings with different sizes. Through extensive experiments on the WHU, Massachusetts, and Inria building datasets, MDTrans achieves intersection over union (IoU) scores of 91.36%, 64.69%, and 79.25%, respectively, outperforming other state‐of‐the‐art models. [ABSTRACT FROM AUTHOR]

As essential components of human society, buildings serve a multitude of functions and significance. Convolutional Neural Network (CNN) has made remarkable progress in the task of building extraction from detailed satellite imagery, owing to the potent capability to capture local information. However, CNN performs suboptimal in extracting larger buildings. Conversely, Transformer has excelled in capturing global information through self-attention mechanisms but are less effective in capturing local information compared to CNN, resulting in suboptimal performance in extracting smaller buildings. Therefore, we have designed the hybrid model STransU2Net, which combines meticulously designed Transformer and CNN to extract buildings of various sizes. In detail, we designed a Bottleneck Pooling Block (BPB) to replace the conventional Max Pooling layer during the downsampling phase, aiming to enhance the extraction of edge information. Furthermore, we devised the Channel And Spatial Attention Block (CSAB) to enhance the target location information during the encoding and decoding stages. Additionally, we added a Swin Transformer Block (STB) at the skip connection location to enhance the model's global modeling ability. Finally, we empirically assessed the performance of STransU2Net on both the Aerial imagery and Satellite II datasets, The IoU achieved state-of-the-art results with 91.04% and 59.09%, respectively, outperforming other models. [ABSTRACT FROM AUTHOR]