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 provide 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, DcapsulesNet (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.
Highlights
D UE to the rapid advancement in imaging technology, remote sensing images could be available more conveniently and economically nowadays and play an essential role in many areas such as military, agriculture [1], and environment monitoring [2], change detection [3], urban planning, land resource management, disaster monitoring, traffic control, and so on [4]–[8]
Due to the complexity of remote sensing images and the loss of deep detailed after the pooling operations in these scene classification methods, the traditional unsupervised feature encoding methods are directly used to integrate all the information of the multilayer convolutional features, which cannot meet the need of generating the discriminative representation of complex scenes
In order to evaluate the performance of our proposed D-CapsNet for scene classification, we conduct the experiments on three benchmark datasets and is compared DCapsNet with state-of-the-art scene classification methods, including CaffeNet [2], Deep CNN Transfer [54], GoogLeNet [2], VGG-16 [2], SAL-PTM [18], mid-level visual dictionary learning [19], SRSCNN [27], fusion by addition [29], two-stream fusion [30], D-CNN [35], FACNN [36], GBNet [37], multifeature fusion ELM [38], bidirectional adaptive feature fusion [39], pyramid multisubset weighted multideep feature fusion (PMWMFF) [40], multiscale triplet loss [41], LGFBOVW [66], finetuned GoogLeNet [67], DSFATN [68], GCFs +LOFs [69], salM3LBP-CLM [70], TEX-Net-LF [71], CNN-ELM [72], CCP-Net [73], and triple networks [74]
Summary
D UE to the rapid advancement in imaging technology, remote sensing images could be available more conveniently and economically nowadays and play an essential role in many areas such as military, agriculture [1], and environment monitoring [2], change detection [3], urban planning, land resource management, disaster monitoring, traffic control, and so on [4]–[8]. With the rapid development and breakthrough of deep learning in computer vision [20]–[24], high-level features from deep learning have been proven to contain both semantic and abstract representations, and are much better and more robust than handcrafted features They have outstanding performance in many applications, such as object detection [20], action recognition [21], image recognition [22], semantic segmentation [23], nature images classification [24]. RAZA et al.: DIVERSE CAPSULES NETWORK COMBINING MULTICONVOLUTIONAL LAYERS significantly reduces the computational and time complexity, and generates discriminative features with effective capsule-level attributes to further reduce the influence of intraclass diversity and high interclass similarity in scene classification.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
