Scholarlink

<p>LULC has a pivotal influence in modeling and tracking changes in ecosystems. This paper focuses on providing a Light CNN architecture built on top of VGG16 with a fuzzy classifier for LULC classification of multispectral satellite images. The methodologies applied are VGG16 with instance normalization and a fuzzy classifier. While the former is used as the backbone network, the latter significantly enhances the quality of the resultant classification maps. With an aim of developing a distinctive model, the issue of unreliability in raw data is addressed by integrating the principles of Fuzzy Logic with the streamlined architecture of Light CNN. The commandability of the proposed model is achieved by comparing its classification accuracy with other standard models, such as AlexNet, VGG16, DenseNet and ResNet50. The experiments are conducted using the Wuhan dense labeling dataset and the Gaofen image dataset to classify LULC into six class labels, with data obtained from China's Gaofen-1 and Gaofen-2 RS satellites. The architecture with an ensemble of fuzzified light CNN outperforms the deep neural network models in terms of performance, allowing for more accurate differentiation with desired accuracies over 84% for all LULC categories (water, forest, buildings, farmland, meadow, and others).</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 lightweight CNN for LULC classification: an alpha-cut fuzzy geometric normalization approach

  • 期刊论文

  • 发表日期：

    2025年6月19日

查看全文
  • V. Anitha,
  • S. Kalaiselvi,
  • V. Gomathi,
  • D. Manimegalai

摘要

LULC has a pivotal influence in modeling and tracking changes in ecosystems. This paper focuses on providing a Light CNN architecture built on top of VGG16 with a fuzzy classifier for LULC classification of multispectral satellite images. The methodologies applied are VGG16 with instance normalization and a fuzzy classifier. While the former is used as the backbone network, the latter significantly enhances the quality of the resultant classification maps. With an aim of developing a distinctive model, the issue of unreliability in raw data is addressed by integrating the principles of Fuzzy Logic with the streamlined architecture of Light CNN. The commandability of the proposed model is achieved by comparing its classification accuracy with other standard models, such as AlexNet, VGG16, DenseNet and ResNet50. The experiments are conducted using the Wuhan dense labeling dataset and the Gaofen image dataset to classify LULC into six class labels, with data obtained from China's Gaofen-1 and Gaofen-2 RS satellites. The architecture with an ensemble of fuzzified light CNN outperforms the deep neural network models in terms of performance, allowing for more accurate differentiation with desired accuracies over 84% for all LULC categories (water, forest, buildings, farmland, meadow, and others).