Scholarlink

<p>Remote detecting maps land use and land cover to create multitemporal images and a concise view. It alludes to arranging human activities, and natural scene components are given acknowledged scientific strategies, like vegetation, metropolitan development, water, and uncovered soil. The profound learning model can deal with grouping land cover classes from colossal volumes of information. This study addresses the challenge of classifying land cover from Sentinel-2 satellite images using a novel residual feature fusion-based Hyper Graph Convolutional Neural Network (HGCNN). This structure is based on Sentinel-2 satellite pictures, incorporating 27,000 clarified and geo-referred images across 13 spectral channels and 10 classes. Two times, leftover elements are disengaged from the pictures as component vectors. Deep-twice residual features are extracted from the images as feature vectors. Each feature vector forms a graph, and these two vector features are fused to form hyper-edges. Hypergraph Convolution Neural Networks with three layers are developed and subsequently prepared by the combined component vector by taking advantage of authentic gaining from high-order highlights. As indicated by exploratory outcomes, the proposed HGCNN technique works better than other cutting-edge strategies and accomplishes a precision level of 99.06%.</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

Remote Sensing Based Land Cover Classification Using Residual Feature—Hyper Graph Convolutional Neural Network (HGCNN)

  • 期刊论文

  • 发表日期：

    2025年1月13日

查看全文
  • L. Gowri,
  • K. R. Manjula,
  • N. Sasikaladevi,
  • S. Pradeepa,
  • Rengarajan Amirtharajan

摘要

Remote detecting maps land use and land cover to create multitemporal images and a concise view. It alludes to arranging human activities, and natural scene components are given acknowledged scientific strategies, like vegetation, metropolitan development, water, and uncovered soil. The profound learning model can deal with grouping land cover classes from colossal volumes of information. This study addresses the challenge of classifying land cover from Sentinel-2 satellite images using a novel residual feature fusion-based Hyper Graph Convolutional Neural Network (HGCNN). This structure is based on Sentinel-2 satellite pictures, incorporating 27,000 clarified and geo-referred images across 13 spectral channels and 10 classes. Two times, leftover elements are disengaged from the pictures as component vectors. Deep-twice residual features are extracted from the images as feature vectors. Each feature vector forms a graph, and these two vector features are fused to form hyper-edges. Hypergraph Convolution Neural Networks with three layers are developed and subsequently prepared by the combined component vector by taking advantage of authentic gaining from high-order highlights. As indicated by exploratory outcomes, the proposed HGCNN technique works better than other cutting-edge strategies and accomplishes a precision level of 99.06%.