Attention-Aware Deep Feature Embedding for Remote Sensing Image Scene Classification

Abstract

Due to the wide application of Remote Sensing (RS) image scene classification, more and more scholars activate great attention to it. With the development of the Convolutional Neural Network (CNN), the CNN-based methods of RS image scene classification have made impressive progress. In the existing works, most of the architectures just considered the global information of the RS images. However, the global information contains a large number of redundant areas that diminish the classification performance and ignore the local information that reflects more fine spatial details of local objects. Furthermore, most CNN-based methods assign the same weights to each feature vector causing the mode to fail to discriminate the crucial features. In this paper, a novel method by Two-branch Deep Feature Embedding with a Dual Attention-Aware module for RS image scene classification is proposed. In order to mine more complementary information, we extract global semanticbased features of high level and local object-based features of low level by the Two-branch Deep Feature Embedding (TDFE) module. Then, to focus selectively on the key global-semantics feature maps as well as the key local regions, we propose a Dual Attention-Aware (DAA) module to attain those key information. We conduct extensive experiments to verify the superiority of our proposed method, and the experimental results obtained on two widely used RS scene classification benchmarks demonstrate the effectiveness of the proposed method.

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 recent works, many researchers have introduced attention into CNN-based RSSC, aiming to improve the RSSC performance. [25][26][27][28][29][30][31][32][33][34] For example, in Ref. 25 Currently, many researchers have attempted to apply the above transformers to the RS SC task.…”

“…In recent works, many researchers have introduced attention into CNN-based RSSC, aiming to improve the RSSC performance 25 – 34 For example, in Ref. 25, Bi et al.…”

“…In Ref. 34, Chen et al. designed a model via two-branch deep feature embedding and a dual attention-aware (TDFE-DAA) for RSSC.…”

A Dense Residual Correlation Attention Multi-scale Network for Remote Sensing Scene Classification

“…In order to fully verify the progress of our proposed method, we compared it with some state-of-the-art methods, including AlexNet [56], GoogleNet [34], CaffeNet [56], VGG-VD-16 [54], TEXNet [20], VGG16-CapsNet [57], VGG-VD-16-SAFF [58], ResNet-LGFFE [59], CSDS [60], MSRes-SplitNet [15], EFPN-DSE [49], TDFE-DAA [61], RANet [52] and EFPN-DSE-TDFF [49]. .…”