Scholarlink

<p>The process of segmentation of hyperspectral images involves dividing the image into various categories by interpreting the spectral data which in turn enhances the accuracy of environmental, urban, and agricultural monitoring. Most existing land-cover land-use segmentation models have operational inefficiencies due to limitations such as high memory consumption, complicated computing processes, and weak scalability with the architectural complexities hindering the performance across different datasets and tasks. In response to this, a Multi Attention Convolutional sparse coding U-network Feed-forward Network (MACUFN) is developed as a multi-layered and advanced solution for segmentation. The Gaussian Adaptive Bilateral Median Filtering (GABMF) is very effective at noise reduction while still maintaining sharp edges. Moreover, the Adaptive Particle Swarm Dandelion Optimization (APSDO) technique underpins the proposed segmentation model’s parameters processing, which allows for proper management of hyperspectral images in different datasets enhancing precision and consistency beyond the existing approaches. The proposed system demonstrates remarkable performance, achieving accuracy across all the datasets as 99.97%, 99.98%, 99.94%, and 99.35% on Indian Pines, Pavia University, Salinas, and EuroSat respectively. The proposed model enhances the accuracy as well as the efficiency of land-cover segmentation and overcomes the drawbacks faced by current land-cover segmentation models.</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

Multi Attention Convolutional Sparse Coding U-Net for Enhanced Land-Use and Land-Cover Segmentation Using Hyperspectral Images

  • 期刊论文

  • 发表日期：

    2025年5月20日

查看全文
  • Vijaykumar P. Yele,
  • Namdeo Baban Badhe,
  • Sujata Alegavi,
  • R. R. Sedamkar

摘要

The process of segmentation of hyperspectral images involves dividing the image into various categories by interpreting the spectral data which in turn enhances the accuracy of environmental, urban, and agricultural monitoring. Most existing land-cover land-use segmentation models have operational inefficiencies due to limitations such as high memory consumption, complicated computing processes, and weak scalability with the architectural complexities hindering the performance across different datasets and tasks. In response to this, a Multi Attention Convolutional sparse coding U-network Feed-forward Network (MACUFN) is developed as a multi-layered and advanced solution for segmentation. The Gaussian Adaptive Bilateral Median Filtering (GABMF) is very effective at noise reduction while still maintaining sharp edges. Moreover, the Adaptive Particle Swarm Dandelion Optimization (APSDO) technique underpins the proposed segmentation model’s parameters processing, which allows for proper management of hyperspectral images in different datasets enhancing precision and consistency beyond the existing approaches. The proposed system demonstrates remarkable performance, achieving accuracy across all the datasets as 99.97%, 99.98%, 99.94%, and 99.35% on Indian Pines, Pavia University, Salinas, and EuroSat respectively. The proposed model enhances the accuracy as well as the efficiency of land-cover segmentation and overcomes the drawbacks faced by current land-cover segmentation models.