Multiscale Short and Long Range Graph Convolutional Network for Hyperspectral Image Classification

Zhu, Wenchao; Zhao, Chunhui; Feng, Shou; Qin, Boao

doi:10.1109/tgrs.2022.3199467

Cited by 23 publications

(9 citation statements)

References 61 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our MSGCRN is implemented with the TensorFlow framework on a Windows server with an 8-core Xeon CPU and a main memory of 16 Giga bytes. The momentum of the BN layer in GCRN is set to 0.9 32 . We use the Adam optimizer with a learning rate of 0.001 and a dynamic adjustment rate of the learning rate of 0.001, and the dynamic update period is 50.…”

Section: Resultsmentioning

confidence: 99%

“…The momentum of the BN layer in GCRN is set to 0.9. 32 We use the Adam optimizer with a learning rate of 0.001 and a dynamic adjustment 1-3. In addition, the experiments were evaluated using three performance metrics, namely overall accuracy (OA), average accuracy (AA), and Kappa coefficient 25 (Kappa), defined as follows:…”

Section: Methodsmentioning

confidence: 99%

“…The network described in Ref. 32 utilizes a combination of multiscale, short/long-distance information, and graph convolutional layers to process HSI data. At the same time, a multiscale processing mechanism is introduced to obtain different levels of features with graph convolution operations at different scales.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Multiscale graph convolution residual network for hyperspectral image classification

Li,

Sun,

Feng

et al. 2024

J. Appl. Rem. Sens.

View full text Add to dashboard Cite

In recent years, graph convolutional networks (GCNs) have attracted increased attention in hyperspectral image (HSI) classification through the utilization of data and their connection graph. However, most existing GCN-based methods have two main drawbacks. First, the constructed graph with pixel-level nodes loses many useful spatial information while high computational cost is required due to large graph size. Second, the joint spatial-spectral structure hidden in HSI are not fully explored for better neighbor correlation preservation, which limits the GCN to achieve promising performance on discriminative feature extraction. To address these problems, we propose a multiscale graph convolutional residual network (MSGCRN) for HSI classification. First, to explore the local spatial-spectral structure, superpixel segmentation is performed on the spectral principal component of HSI at different scales. Thus, the obtained multiscale superpixel areas can capture rich spatial texture division. Second, multiple superpixel-level subgraphs are constructed with adaptive weighted node aggregation, which not only effectively reduces the graph size, but also preserves local neighbor correlation in varying subgraph scales. Finally, a graph convolution residual network is designed for multiscale hierarchical features extraction, which are further integrated into the final discriminative features for HSI classification via a diffusion operation. Moreover, a mini-batch branch is adopted to the large-scale superpixel branch of MSGCRN to further reduce computational cost. Extensive experiments on three public HSI datasets demonstrate the advantages of our MSGCRN model compared to several cutting-edge approaches.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Multiscale graph convolution residual network for hyperspectral image classification

Li,

Sun,

Feng

et al. 2024

J. Appl. Rem. Sens.

View full text Add to dashboard Cite

show abstract

“…Most of these methods use deep neural networks for remote sensing data classification. Multiscale feature fusion [4], dual-branch network [12], hypergraph convolution [13] [14], Transformer-like network [15], attention multihop graph [16], and neural architecture search [17] have been used for hyperspectral or SAR data feature extraction. Although impressive results have been achieved, single modality classification can be problematic in some cases.…”

Section: Introductionmentioning

confidence: 99%

SS-MAE: Spatial–Spectral Masked Autoencoder for Multisource Remote Sensing Image Classification

Lin,

Gao,

Shi

et al. 2023

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Masked image modeling (MIM) is a highly popular and effective self-supervised learning method for image understanding. Existing MIM-based methods mostly focus on spatial feature modeling, neglecting spectral feature modeling. Meanwhile, existing MIM-based methods use Transformer for feature extraction, some local or high-frequency information may get lost. To this end, we propose a spatial-spectral masked auto-encoder (SS-MAE) for HSI and LiDAR/SAR data joint classification. Specifically, SS-MAE consists of a spatial-wise branch and a spectral-wise branch. The spatial-wise branch masks random patches and reconstructs missing pixels, while the spectral-wise branch masks random spectral channels and reconstructs missing channels. Our SS-MAE fully exploits the spatial and spectral representations of the input data. Furthermore, to complement local features in the training stage, we add two lightweight CNNs for feature extraction. Both global and local features are taken into account for feature modeling. To demonstrate the effectiveness of the proposed SS-MAE, we conduct extensive experiments on three publicly available datasets. Extensive experiments on three multi-source datasets verify the superiority of our SS-MAE compared with several state-of-the-art baselines. The source codes are available at https://github.com/summitgao/SS-MAE.

show abstract

“…Recently, Zhu et al proposed a multi-scale shortand long-range graph convolutional network (MSLGCN) for HSIC. Multi-scale spatial embeddings and global spectral features are deeply explored by an elegant multi-stage structure [40]. Third, holistic image representation can be better learned by collecting second-order statistics of convolutional features.…”

Section: Introductionmentioning

confidence: 99%

Low-Rank Constrained Attention-Enhanced Multiple Spatial–Spectral Feature Fusion for Small Sample Hyperspectral Image Classification

Feng¹,

Zhang²,

Zhang³

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Hyperspectral images contain rich features in both spectral and spatial domains, which bring opportunities for accurate recognition of similar materials and promote various fine-grained remote sensing applications. Although deep learning models have been extensively investigated in the field of hyperspectral image classification (HSIC) tasks, classification performance is still limited under small sample conditions, and this has been a longstanding problem. The features extracted by complex network structures with large model size are redundant to some extent and prone to overfitting. This paper proposes a low-rank constrained attention-enhanced multiple feature fusion network (LAMFN). Firstly, factor analysis is used to extract very few components that can describe the original data using covariance information to perform spectral feature preprocessing. Then, a lightweight attention-enhanced 3D convolution module is used for deep feature extraction, and the position-sensitive information is supplemented using a 2D coordinate attention. The above widely varying spatial–spectral feature groups are fused through a simple composite residual structure. Finally, low-rank second-order pooling is adopted to enhance the convolutional feature selectivity and achieve classification. Extensive experiments were conducted on four representative hyperspectral datasets with different spatial–spectral characteristics, namely Indian Pines (IP), Pavia Center (PC), Houston (HU), and WHU-HongHu (WHU). The contrast methods include several advanced models proposed recently, including residual CNNs, attention-based CNNs, and transformer-based models. Using only five samples per class for training, LAMFN achieved overall accuracies of 78.15%, 97.18%, 81.35%, and 87.93% on the above datasets, which has an improvement of 0.82%, 1.12%, 1.67%, and 0.89% compared to the second-best model. The running time of LAMFN is moderate. For example, the training time of LAMFN on the WHU dataset was 29.1 s, and the contrast models ranged from 3.0 s to 341.4 s. In addition, ablation experiments and comparisons with some advanced semi-supervised learning methods further validated the effectiveness of the proposed model designs.

show abstract

Multiscale Short and Long Range Graph Convolutional Network for Hyperspectral Image Classification

Cited by 23 publications

References 61 publications

Multiscale graph convolution residual network for hyperspectral image classification

Multiscale graph convolution residual network for hyperspectral image classification

SS-MAE: Spatial–Spectral Masked Autoencoder for Multisource Remote Sensing Image Classification

Low-Rank Constrained Attention-Enhanced Multiple Spatial–Spectral Feature Fusion for Small Sample Hyperspectral Image Classification

Contact Info

Product

Resources

About