Scholarlink

<p>Hyperspectral Image (HSI) classification stands as a significant area of research within the realm of remote sensing. Hyperspectral remote sensing image data provide detailed information about ground objects. However, it is difficult to visualize since HSI has high dimensionality, and it leads to larger data files, which can increase storage requirements and complicate data management. Hence, a novel Patch projection matrix + LeNet approach is devised for dimensionality reduction in HSI classification. The input hyperspectral image is acquired from the remote sensing HSI datasets of the Indian Pines and University of Pavia. Subsequently, the HSI matrix is done utilizing the intraclass hyper Laplacian matrix, interclass hyper Laplacian matrix, supervised locality Laplacian matrix, weighted neighborhood margin scattered matrix, and patch matrix. Then, the patch projection matrix is created using a Convolutional Neural Network (CNN). Finally, HSI classification will be conducted using LeNet. A comparative analysis is performed using Principal Component Analysis (PCA), Joint Sparse Local Linear Discriminant Analysis (JSLLDA), Singular Value Decomposition (SVD), Supervised Fractal Dimension Reduction (SFDR), Locally Linear Embedding (LLE), Lightweight-VGG (LVGG), Uniform Manifold Approximation and Projection (UMAP), and Independent Component Analysis (ICA), in conjunction with recent models. Thus, the Patch projection matrix + LeNet model effectively obtained the highest value in outlier samples with an accuracy of 89.86%, True Positive Rate (TPR) of 88.60% and True Negative Rate (TNR) of 90.58%, respectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A patch projection matrix for dimensionality reduction in hyperspectral image classification

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Chandra Sekhar Kintali,
  • Durairaj Latha

摘要

Hyperspectral Image (HSI) classification stands as a significant area of research within the realm of remote sensing. Hyperspectral remote sensing image data provide detailed information about ground objects. However, it is difficult to visualize since HSI has high dimensionality, and it leads to larger data files, which can increase storage requirements and complicate data management. Hence, a novel Patch projection matrix + LeNet approach is devised for dimensionality reduction in HSI classification. The input hyperspectral image is acquired from the remote sensing HSI datasets of the Indian Pines and University of Pavia. Subsequently, the HSI matrix is done utilizing the intraclass hyper Laplacian matrix, interclass hyper Laplacian matrix, supervised locality Laplacian matrix, weighted neighborhood margin scattered matrix, and patch matrix. Then, the patch projection matrix is created using a Convolutional Neural Network (CNN). Finally, HSI classification will be conducted using LeNet. A comparative analysis is performed using Principal Component Analysis (PCA), Joint Sparse Local Linear Discriminant Analysis (JSLLDA), Singular Value Decomposition (SVD), Supervised Fractal Dimension Reduction (SFDR), Locally Linear Embedding (LLE), Lightweight-VGG (LVGG), Uniform Manifold Approximation and Projection (UMAP), and Independent Component Analysis (ICA), in conjunction with recent models. Thus, the Patch projection matrix + LeNet model effectively obtained the highest value in outlier samples with an accuracy of 89.86%, True Positive Rate (TPR) of 88.60% and True Negative Rate (TNR) of 90.58%, respectively.