A Multi-Scale Progressive Collaborative Attention Network for Remote Sensing Fusion Classification

Ma, Wenping; Li, Yating; Zhu, Hao; Ma, Haoxiang; Jiao, Licheng; Shen, Jianchao; Hou, Biao

doi:10.1109/tnnls.2021.3121490

Cited by 25 publications

(3 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this experiment, we changed the size of the 2D flattening patch sequence and set the patch size to 16 and 32. According to the literature 20 , the patch size should be 14, 16, and 32, but due to hardware limitation, the experimental setting was limited to the smallest patch size (i.e., 16). The corresponding results, which indicate that the number of linear embedding sequences input to the encoder module is inversely proportional to patch size 20 , are shown in Table 7.…”

Section: Nwpu Datasetmentioning

confidence: 99%

“…According to the literature 20 , the patch size should be 14, 16, and 32, but due to hardware limitation, the experimental setting was limited to the smallest patch size (i.e., 16). The corresponding results, which indicate that the number of linear embedding sequences input to the encoder module is inversely proportional to patch size 20 , are shown in Table 7. The smaller the patch size, the more pieces cut; the larger the patch size, the fewer pieces are cut.…”

Section: Nwpu Datasetmentioning

confidence: 99%

“…Remote sensing images not only contain rich channel information but also spatial information. To fully extract their features, Ma et al 20 proposed an adaptive multi-scale spatial attention module (AMSA) and an adaptive multi-scale channel attention module (AMCA) based on image characteristics by adopting adaptive expansion rate selection strategy (ADR-SS), so as to increase the diversity of extracted features. Zhu et al 21 designed an adaptive spatial attention module (ASA-Module) and an adaptive channel attention module (ACA-Module) to strengthen spatial features from both larger-with smaller-sized targets and spectral features among channels.…”

mentioning

confidence: 99%

See 2 more Smart Citations

Transformer based on channel-spatial attention for accurate classification of scenes in remote sensing image

Guo

Jia

Bai

2022

Sci Rep

View full text Add to dashboard Cite

Recently, the scenes in large high-resolution remote sensing (HRRS) datasets have been classified using convolutional neural network (CNN)-based methods. Such methods are well-suited for spatial feature extraction and can classify images with relatively high accuracy. However, CNNs do not adequately learn the long-distance dependencies between images and features in image processing, despite this being necessary for HRRS image processing as the semantic content of the scenes in these images is closely related to their spatial relationship. CNNs also have limitations in solving problems related to large intra-class differences and high inter-class similarity. To overcome these challenges, in this study we combine the channel-spatial attention (CSA) mechanism with the Vision Transformer method to propose an effective HRRS image scene classification framework using Channel-Spatial Attention Transformers (CSAT). The proposed model extracts the channel and spatial features of HRRS images using CSA and the Multi-head Self-Attention (MSA) mechanism in the transformer module. First, the HRRS image is mapped into a series of multiple planar 2D patch vectors after passing to the CSA. Second, the ordered vector is obtained via the linear transformation of each vector, and the position and learnable embedding vectors are added to the sequence vector to capture the inter-feature dependencies at a distance from the generated image. Next, we use MSA to extract image features and the residual network structure to complete the encoder construction to solve the gradient disappearance problem and avoid overfitting. Finally, a multi-layer perceptron is used to classify the scenes in the HRRS images. The CSAT network is evaluated using three public remote sensing scene image datasets: UC-Merced, AID, and NWPU-RESISC45. The experimental results show that the proposed CSAT network outperforms a selection of state-of-the-art methods in terms of scene classification.

show abstract

Section: Nwpu Datasetmentioning

confidence: 99%