Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spatial-Geometry Enhanced 3D Dynamic Snake Convolutional Neural Network for Hyperspectral Image Classification

Guandong Li, Mengxia Ye

TL;DR

This work tackles hyperspectral image classification under spatial variability and spectral redundancy by introducing Spatial-Geometry Enhanced 3D Dynamic Snake Network (SG-DSCNet). The method combines Dynamic Snake Convolution (DSConv) with a multi-view fusion strategy on a modified 3D-DenseNet, enabling deformable yet constrained kernel adaptation and diverse geometric observations. DSConv leverages deformable offsets within a $9×9$ receptive field and axis-wise accumulation, while the multi-view templates $T_l$ are fused with dropout to boost robustness. Across Indian Pines, Pavia University, and KSC, SG-DSCNet achieves state-of-the-art OA, AA, and Kappa with efficient computation, demonstrating improved spatial-spectral representation without extra network depth or width.

Abstract

Deep neural networks face several challenges in hyperspectral image classification, including complex and sparse ground object distributions, small clustered structures, and elongated multi-branch features that often lead to missing detections. To better adapt to ground object distributions and achieve adaptive dynamic feature responses while skipping redundant information, this paper proposes a Spatial-Geometry Enhanced 3D Dynamic Snake Network (SG-DSCNet) based on an improved 3D-DenseNet model. The network employs Dynamic Snake Convolution (DSCConv), which introduces deformable offsets to enhance kernel flexibility through constrained self-learning, thereby improving regional perception of ground objects. Additionally, we propose a multi-view feature fusion strategy that generates multiple morphological kernel templates from DSCConv to observe target structures from different perspectives and achieve efficient feature fusion through summarizing key characteristics. This dynamic approach enables the model to focus more flexibly on critical spatial structures when processing different regions, rather than relying on fixed receptive fields of single static kernels. The DSC module enhances model representation capability through dynamic kernel aggregation without increasing network depth or width. Experimental results demonstrate superior performance on the IN, UP, and KSC datasets, outperforming mainstream hyperspectral classification methods.

Spatial-Geometry Enhanced 3D Dynamic Snake Convolutional Neural Network for Hyperspectral Image Classification

TL;DR

This work tackles hyperspectral image classification under spatial variability and spectral redundancy by introducing Spatial-Geometry Enhanced 3D Dynamic Snake Network (SG-DSCNet). The method combines Dynamic Snake Convolution (DSConv) with a multi-view fusion strategy on a modified 3D-DenseNet, enabling deformable yet constrained kernel adaptation and diverse geometric observations. DSConv leverages deformable offsets within a receptive field and axis-wise accumulation, while the multi-view templates are fused with dropout to boost robustness. Across Indian Pines, Pavia University, and KSC, SG-DSCNet achieves state-of-the-art OA, AA, and Kappa with efficient computation, demonstrating improved spatial-spectral representation without extra network depth or width.

Abstract

Deep neural networks face several challenges in hyperspectral image classification, including complex and sparse ground object distributions, small clustered structures, and elongated multi-branch features that often lead to missing detections. To better adapt to ground object distributions and achieve adaptive dynamic feature responses while skipping redundant information, this paper proposes a Spatial-Geometry Enhanced 3D Dynamic Snake Network (SG-DSCNet) based on an improved 3D-DenseNet model. The network employs Dynamic Snake Convolution (DSCConv), which introduces deformable offsets to enhance kernel flexibility through constrained self-learning, thereby improving regional perception of ground objects. Additionally, we propose a multi-view feature fusion strategy that generates multiple morphological kernel templates from DSCConv to observe target structures from different perspectives and achieve efficient feature fusion through summarizing key characteristics. This dynamic approach enables the model to focus more flexibly on critical spatial structures when processing different regions, rather than relying on fixed receptive fields of single static kernels. The DSC module enhances model representation capability through dynamic kernel aggregation without increasing network depth or width. Experimental results demonstrate superior performance on the IN, UP, and KSC datasets, outperforming mainstream hyperspectral classification methods.

Paper Structure

This paper contains 20 sections, 8 equations, 9 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Spatial-Geometry Enhanced 3D Dynamic Snake CNN
  3. Dynamic Convolution
  4. 3D Snake Convolution and Multi-View Fusion
  5. Dynamic Snake Convolution
  6. Multi-View Fusion
  7. 3D-CNN Framework for HSI Feature Extraction
  8. Exponentially Increasing Growth Rate
  9. Fully Dense Connectivity
  10. Experiments and Analysis
  11. Datasets
  12. Indian Pines Dataset
  13. Pavia University Dataset
  14. KSC Dataset
  15. Experimental Analysis
  16. ...and 5 more sections

Figures (9)

  • Figure 1: Comparison between standard convolution, dilated convolution, deformable convolution, and DSC Conv
  • Figure 2: DSCConv architecture diagram
  • Figure 3: Multi-view fusion strategy diagram
  • Figure 4: The proposed DenseNet variant. It differs from the original DenseNet in two aspects: (1) layers with different resolution feature maps are also directly connected; (2) the growth rate doubles whenever the feature map size is downsampled (features generated in the third yellow dense block are significantly more numerous than those in the first block).
  • Figure 5: SG-DSCNet architecture incorporating modified 3D-DenseNet framework
  • ...and 4 more figures