High Resolution SAR Image Classification Using Global-Local Network Structure Based on Vision Transformer and CNN

Liu, Xingyu; Wu, Yan; Liang, Wenkai; Cao, Yice; Li, Ming

doi:10.1109/lgrs.2022.3151353

Cited by 25 publications

(15 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, we introduce a classification layer after the hash network layer to improve the feature extraction capability and at the same time enhance the class discrimination capability of the model. In single-label image classification scenarios [45][46], we often use the classical cross-entropy loss (CE), which is denoted as:…”

Section: B Hash Learningmentioning

confidence: 99%

ICIHRN: An Interpretable Multilabel Hash Retrieval Method for Satellite Cloud Images

Jin,

Cai,

Pan

et al. 2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Observing clouds to understand the weather is a crucial method for people to forecast upcoming conditions. Utilizing content-based satellite cloud image retrieval allows for the swift discovery of comparable historical cloud images, significantly aiding meteorologists in their advanced investigations. Nevertheless, satellite cloud images often present complexities due to their inclusion of diverse cloud types, leading to inadequate retrieval outcomes when relying on conventionally employed single-label retrieval techniques. Despite notable accomplishments in cloud image retrieval applications utilizing deep neural networks, concerns regarding network interpretability undermine confidence in the model's deductive outcomes. This paper introduces the interpretable cloud image hash retrieval network (ICIHRN), a framework that employs a singular object-level global unit alongside multiple local feature units for the purpose of generating hash codes tailored to cloud image retrieval. Furthermore, an attention branching network is incorporated to enhance the model's focus on discriminative regions within the image. Additionally, a suppression module is implemented to progressively uncover complementary regions through the suppression of prominent areas in preceding layers and the amalgamation of relationships among activated regions. This ensures that each feature unit is endowed with distinctive semantic information, thereby imparting a level of interpretability to the retrieval outcomes. On this foundation, multi-label supervision is seamlessly integrated into the deep hash learning framework. This integration not only enhances the depiction of intricate semantic contents within cloud images but also boosts retrieval efficiency. Comprehensive experimental outcomes, grounded in the publicly accessible satellite cloud map dataset LSCIDMR-V2, demonstrate superior performance relative to other methods.

show abstract

Section: B Hash Learningmentioning

confidence: 99%

ICIHRN: An Interpretable Multilabel Hash Retrieval Method for Satellite Cloud Images

Jin,

Cai,

Pan

et al. 2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…But these models have higher complexity, which limits their applicability. To overcome this issue, work in [7] proposes design of global-local network structure with CNN, which assists in improving classification accuracy for multiple use cases. It uses vision transformers, which makes it useful for high density feature extraction & classification deployments.…”

Section: Literature Reviewmentioning

confidence: 99%

“…2. Some techniques that are effective in identifying synthetic images require significant computational resources, which limits their scalability for real-world applications [7,8,10]. 3.…”

Section: Issues With Existing Techniquesmentioning

confidence: 99%

Design of a Hybrid GWO CNN Model for Identification of Synthetic Images via Transfer Learning Process

2023

IJIES

View full text Add to dashboard Cite

Visual representation of synthetic images is very accurate, due to which it is difficult to differentiate them for their natural counterparts. Existing models that perform this differentiation are either very complex, or cannot be scaled for multidomain image sets. Moreover, the accuracy of these models depends directly upon type of dataset & feature sets used for training & validation purposes. To overcome these limitations, this paper proposes design of a Hybrid GWO CNN Model for identification of Synthetic Images for big data Applications. The proposed model initially extracts multidomain feature sets from input images, that includes wavelet, cosine, fourier & convolutional features. These features are processed via a grey wolf optimization (GWO) Model, that assists in improving inter-class feature variance while minimizing intra-class variance levels. The GWO Model identifies training & validation sets, thereby assisting the classification model to accurately differentiate between different image types. To perform this task, a variance-based fitness function was modelled that covers both inter-class & inter-class variance levels. This classification is performed via use of a transfer learning-based CNN Model, that extends VGG-19 for high-efficiency operations. The proposed model was tested on a large number of datasets including Synthetic Fruit, Unsplash, ESPL Synthetic Image, and Okazaki Synthetic Texture Image (OSTI) databases. Based on these datasets, accuracy, precision, computational delay, recall & AUC (Area Under the Curve) metrics were evaluated & compared with existing models. It was observed that the proposed model showcased 9.5% better accuracy, 8.3% higher precision, 6.5% better recall, and 3.9% faster performance when compared with existing models. Due to such high performance under different datasets, the proposed model is useful for large-scale synthetic-image identification use cases.

show abstract

“…ViT 31 has successfully applied the transformer 30 from natural language processing to computer vision 32 – 34 It demonstrates that transformer also has a strong ability to model spatial correlations of images. The main reason for the success of the transformer is its core component, the self-attention module that captures global information.…”

Section: Related Workmentioning

confidence: 99%

Synthetic aperture radar and optical image registration using local and global feature learning by modality-shared attention network

Hu,

Wu,

et al. 2023

J. Appl. Rem. Sens.

View full text Add to dashboard Cite

High Resolution SAR Image Classification Using Global-Local Network Structure Based on Vision Transformer and CNN

Cited by 25 publications

References 12 publications

ICIHRN: An Interpretable Multilabel Hash Retrieval Method for Satellite Cloud Images

ICIHRN: An Interpretable Multilabel Hash Retrieval Method for Satellite Cloud Images

Design of a Hybrid GWO CNN Model for Identification of Synthetic Images via Transfer Learning Process

Synthetic aperture radar and optical image registration using local and global feature learning by modality-shared attention network

Contact Info

Product

Resources

About