SCAttNet: Semantic Segmentation Network With Spatial and Channel Attention Mechanism for High-Resolution Remote Sensing Images

Abstract

High-resolution remote sensing images (HRRSIs) contain substantial ground object information, such as texture, shape, and spatial location. Semantic segmentation, which is an important method for element extraction, has been widely used in processing mass HRRSIs. However, HRRSIs often exhibit large intraclass variance and small interclass variance due to the diversity and complexity of ground objects, thereby bringing great challenges to a semantic segmentation task. In this study, we propose a new end-to-end semantic segmentation network, which integrates two lightweight attention mechanisms that can refine features adaptively. We compare our method with several previous advanced networks on the ISPRS Vaihingen and Potsdam datasets. Experimental results show that our method can achieve better semantic segmentation results compared with other works. The source codes are available at https://github.com/lehaifeng/SCAttNet.

Land-Use Mapping for High-Spatial Resolution Remote Sensing Image Via Deep Learning: A Review

“…The increasing number of HSR-RSIs enable building largescale segmentation datasets that play an indispensable part in advance of semantic segmentation. In the past few years, several publicly available HSR-RSIs benchmark datasets have been proposed by different research groups for LUM of remote sensing images [26]- [35].…”

“…We split the labeled images of the Vaihingen dataset into a training dataset (12 images of ID 1,3,5,7,11,13,15,17,21,23,26,28) and a test dataset (4 images of ID 30,32,34,37). We randomly crop the training images into a size of 256 × 256 and flip and rotation images for data augmentation.…”

DSPCANet: Dual-Channel Scale-Aware Segmentation Network With Position and Channel Attentions for High-Resolution Aerial Images

“…Misra et al [47] proposed a method to compute the attention weights by capturing the crossdimension interaction using a three-branch structure. In SCAt-tNet V2 [48], both channel and space attention modules were utilized to adaptively refine the extracted features of remote sensing images. In addition, by considering the geometrical structure information of remote sensing data, a spatial-spectral second-order attention module, a multiscale residual attention module, and a tensor attention module can produce satisfactory classification results in data fusion and classification [49], [50].…”

“…Moreover, in the U-Net [10] and FCN 32s models, VGG-16 [52] is selected as the backbone. For RefineNet [38], Deeplabv3+ [37], PSPNet [17], SCAttNet V2 [48], ResUNet-a [14], and CBAM [56], ResNet50 [53] is used as the backbone, which is downsampled 32 times for RefineNet, 8 times for Deeplabv3+, 8 times for SCAttNet V2, 8 times for PSPNet, and 8 times for ResUNet-a, while 8 times for CBAM. In DSPCANet (IRRG), only the IRRG images are considered.…”

Adaptive Filtering Remote Sensing Image Segmentation Network based on Attention Mechanism

“…Unlike previous works that capture contexts by multi-scale feature fusion, Dual Attention Network (DANet) [19] adaptively integrate local features with their global dependencies, which model the semantic interdependencies in spatial and channel dimensions respectively. SCAttNet [20] proposed by Haifeng Li combines spatial attention with channel attention to segment remote sensing imagery. In summary, the following problems still exist when using deep learning networks to segment remote sensing imagery:…”