Diverse Capsules Network Combining Multiconvolutional Layers for Remote Sensing Image Scene Classification

Abstract

Remote sensing image scene classification has drawn significant attention for its potential applications in the economy and livelihoods. Unlike the traditional handcrafted features, the convolutional neural networks (CNNs) provides an excellent avenue in obtaining powerful discriminative features. Although tremendous efforts have been made so far in this domain, there are still many open challenges in scene classification due to the scene complexity with higher within-class diversity and between-class similarity. To solve the above-mentioned problems, D-CapsNet is proposed to learn the richer and more robust features for scene classification. It is an end to end network with four types of layers and incorporates visual attention mechanisms. Its diverse capsules encode different properties of complex image scenes, including deep high-level features, spatial attention based on the fusion of multilayers features, both spatial and channel attention based on high-level features, and their fusion. Experiments on three image scene datasets demonstrate that D-CapsNet outperforms other baselines and state-of-the-art methods with a significant improvement in both classification accuracy and speed.

Deep Object-Centric Pooling in Convolutional Neural Network for Remote Sensing Scene Classification

“…Compare with AlexNet [53], VGG-VD16 [53], and Scenario (II) [10], which only uses the original pre-trained CNN model, DOCP-net manages a higher classification performance due to combining object-level and scene-level features. The proposed method obtains comparable results with D-CapsNet [54] and VGG-VD16+CapsNet [55]. According to Fig.…”

“…This model may have an advantage when the data set's scale increases. In [60], a self-attention- [10] 96.90 (0.77) AlexNet+SPP [17] 95.95 (1.01) CCP-net [18] 97.52 (0.97) AlexNet+MSCP [56] 97.29 (0.63) VGG-VD16+MSCP [56] 98.36 (0.58) BAFF [57] 95.48 (0.22) RADC-Net [58] 97.05 (0.48) D-CapsNet [54] 99.05 (0.12) DDRL-AM [59] 99.05 (0.08) VGG-VD16+CapsNet [55] 98.81 (0.12) AlexNet+SAFF [60] 96.13 ( 0.97) VGG-VD16+SAFF [60] 97.02 ( 0.78) [18], a concentric circle pooling layer is proposed to incorporate rotation-invariant spatial layout information of remote sensing scene images.…”

“…The DDRL-AM is not a completely end-to-end learning architecture, with the effectiveness of the attention map depending on the outputs of the fine-tuned CNN models. Although the performance of D-CapsNet[54] is better than DOCP, it utilizes spatial and channel attention mechanisms to enhance accuracy. The D-CNN method[62] based on VGG-VD16 achieves better classification accuracies than DOCP, but it highly depends on selecting the input image pair, which can be very time-consuming, and imposes a large memory allocation during the training phase.…”

Embedded Self-Distillation in Compact Multibranch Ensemble Network for Remote Sensing Scene Classification

“…. , N. We average these feature maps to obtain a teacher feature map F H * W e as (10). The student is the main-branch feature map after normalization F H * W m .…”

IR-MSDNet: Infrared and Visible Image Fusion Based On Infrared Features and Multiscale Dense Network

“…The encoder contains different modules for feature extraction, including dense module, multiscale module, feature fusion module, channel-wise attention (CWA) [28] module and fused feature bank. In order to receive a visible or infrared image of arbitrary size, this image is first processed via the convolutional layer, where kernel size is 3×3, and the stride is 1.…”