Relation-Attention Networks for Remote Sensing Scene Classification

Abstract

Remote sensing (RS) scene classification plays an important role in a wide range of RS applications. Recently, convolutional neural networks (CNNs) have been applied to the field of scene classification in RS images and achieved impressive performance. However, to classify RS scenes, most of the existing CNN methods either utilize the high-level features from the last convolutional layer of CNNs, missing much important information existing at the other levels, or directly fuse the features at different levels, bringing redundant and/or mutually exclusive information. Inspired by the attention mechanism of the human visual system, in this paper, we explore a novel relation-attention model and design an end-to-end relationattention network (RANet) to learn powerful feature representations of multiple levels to further improve the classification performance. First, we propose to extract convolutional features at different levels by pre-trained CNNs. Second, a multi-scale feature computation module is constructed to connect features at different levels and generate multi-scale semantic features. Third, a novel relation-attention model is designed to focus on the critical features whilst avoiding the use of redundant and even distractive ones by exploiting the scale contextual information. Finally, the resulting relation-attention features are concatenated and fed into a softmax layer for the final classification. Experiments on four well-known RS scene classification data sets (UC-Merced, WHU-RS19, AID, and OPTIMAL-31) show that our method outperforms some state-ofthe-art algorithms. The code of our proposed method is publicly

Zero-Shot Remote Sensing Scene Classification Method Based on Local-Global Feature Fusion and Weight Mapping Loss

“…To quantitatively evaluate the performance of our method, we have used several quantitative evaluation metrics including overall accuracy (OA), standard deviation (SD), confusion matrix (CM), and class average accuracy (AA) [27], [62]. OA is a direct measure of the classification accuracy of the model on the entire dataset:…”

Multi-attention aggregation network for remote sensing scene classification

“…To further show the effect of our MAANet, we compare it with a set of state-of-the-art RSSC algorithms, covering traditional non-DL methods (i.e., BoVW, 7 IFK, 7 LDA, 7 LLC 8 ) that mainly rely on mid-level features and DL-based methods that are closely related to our network. Specifically, these DL models are subdivided into: (1) traditional CNNs (i.e., GoogLeNet, 7 CaffeNet, 7 VGG-VD-16, 7 and VGG-16-CapsNet 15 ); (2) gated networks (i.e., GBNet 18 and GBNet + global feature 18 ); (3) feature pyramid networks (i.e., EFPN-DSE-TDFF 19 and RANet 20 ); (4) global–local feature fusion networks (i.e., LCNN-BFF, 21 HABFNet, 22 MF2Net, 23 and DAFGCN 24 ); (5) attention-based networks (i.e., MS2AP, 25 MSA-Network, 26 SAFF, 27 ResNet50+EAM, 28 ACNet, 29 CSDS, 30 SEMSDNet, 31 ACR-MLFF, 32 CRAN, 33 and TDFE-DAA); 34 and (6) currently popular transformers (i.e., ViT-B_32, 35 T2T-ViT-12, 36 V16_21k, 37 ViT, 35 PVT-V2-B0, 38 PiT-S, 39 Swin-T, 40 PVT-Medium, 41 and T-CNN 42 ). For a fair comparison, all results are obtained by the source codes or provided by the authors directly.…”

“…In Ref. 20, Wang et al. constructed a relation-attention guided feature pyramid network (RANet) to learn multilevel features.…”

Text guided zero-shot scene classification of high spatial resolution remote sensing images

“…In Ref. 29, Wang et al. combined the relation network and attention mechanism to learn powerful feature representations of multiple levels to further improve the classification performance.…”