Scholarlink

<p>Human-induced changes in land use and land cover (LULC) significantly affect ecosystems and climate systems. Accurate land cover (LC) classification is therefore essential for sustainable urban planning, environmental protection, and climate change mitigation. The fusion of multi-scale extracted features is disregarded by most current approaches, which cannot fully extract geometric structural information. This study presents a unique deep learning-based approach for LC classification using multispectral data. Initially, we evaluate with six phases: data collection, pre-processing, attribute extraction, attribute selection, segmentation and categorization. Then, the collected input images are used to improve the colour corrections, noise reduction, and normalization for better classification outcomes. After that, the relevant low- and high-level features from pre-processed images are extracted using the Modified ResNeXt technique. Select important features by eliminating unnecessary features using the Improved Binary Crayfish Optimization Algorithm (IBCOA). Then, the Residual-attention UNet++ (RA-UNet++) method is applied to segment the changed land areas based on selected features. Finally, we proposed a novel efficient DenseXtBi-LSTM approach, which is the ensemble of DenseXtNet and Enhanced Bi-directional Long Short-Term Memory (EBi-LSTM) to classify land cover changes with enhanced precision and adaptability accurately. The proposed system outperforms the previously created SOTA segmentation and categorization models with a recall, precision, f1-score, and IoU of over 99% on the test scenes. The quantitative evaluations show that the proposed method works well and quickly for categorizing land cover tasks. According to the study’s findings, a significant change occurred in terms of a decline in greenness due to the fast rise in urbanization, population density, and other infrastructure innovations. As a result, in the upcoming years of rapid environmental change, these findings will be utilized for agricultural management and regional and urban planning.</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 Novel Deep Learning Approach for Multispectral Land Use/Land Cover Classification Using Optimized Feature Selection and Multi-scale Segmentation

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Surendra Reddy Vinta,
  • Sai Srikar Reddy Kolli,
  • Anumula Sruthi

摘要

Human-induced changes in land use and land cover (LULC) significantly affect ecosystems and climate systems. Accurate land cover (LC) classification is therefore essential for sustainable urban planning, environmental protection, and climate change mitigation. The fusion of multi-scale extracted features is disregarded by most current approaches, which cannot fully extract geometric structural information. This study presents a unique deep learning-based approach for LC classification using multispectral data. Initially, we evaluate with six phases: data collection, pre-processing, attribute extraction, attribute selection, segmentation and categorization. Then, the collected input images are used to improve the colour corrections, noise reduction, and normalization for better classification outcomes. After that, the relevant low- and high-level features from pre-processed images are extracted using the Modified ResNeXt technique. Select important features by eliminating unnecessary features using the Improved Binary Crayfish Optimization Algorithm (IBCOA). Then, the Residual-attention UNet++ (RA-UNet++) method is applied to segment the changed land areas based on selected features. Finally, we proposed a novel efficient DenseXtBi-LSTM approach, which is the ensemble of DenseXtNet and Enhanced Bi-directional Long Short-Term Memory (EBi-LSTM) to classify land cover changes with enhanced precision and adaptability accurately. The proposed system outperforms the previously created SOTA segmentation and categorization models with a recall, precision, f1-score, and IoU of over 99% on the test scenes. The quantitative evaluations show that the proposed method works well and quickly for categorizing land cover tasks. According to the study’s findings, a significant change occurred in terms of a decline in greenness due to the fast rise in urbanization, population density, and other infrastructure innovations. As a result, in the upcoming years of rapid environmental change, these findings will be utilized for agricultural management and regional and urban planning.