Scholarlink

<p>The Pichavaram mangrove forest in Tamil Nadu is one of India’s most ecologically significant regions, supporting coastal health and local communities. However, effective mangrove area classification remains challenging due to field inaccessibility and inefficiency of traditional assessment methods, highlighting the demand for advanced solutions. As the existing remote sensing-based studies suffer from limited classification accuracy and high computational complexity, this study combined Hyperspectral Image (HSI) with an Optimized Channel Level Residual CNN (OC-LRCNN) model for improved results in mangrove-related research. The proposed model employs unsupervised feature extraction to capture essential patterns with minimal training data while channel-level residual connections enhance discriminative feature selection and reduce spectral redundancy. Utilizing the Pichavaram EO-1 Hyperion and AVIRIS-NG datasets, the proposed model is compared with traditional CNN, state-of-the-art deep learning architectures (VGG, ResNet, DenseNet) and machine learning methods like SVM and RF. The OC-LRCNN achieved classification accuracies of 98.2% and 99.0% for the Hyperion and AVIRIS-NG datasets with consistently high precision, recall, F1-score and kappa values. These findings demonstrate the model’s effectiveness in reliable mangrove classification and monitoring applications.</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

Mangrove area classification in Pichavaram using Hyperspectral Imaging and Optimized Channel-Level Residual CNN framework

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Ilamathi Parthiban,
  • Chidambaram Somanathan

摘要

The Pichavaram mangrove forest in Tamil Nadu is one of India’s most ecologically significant regions, supporting coastal health and local communities. However, effective mangrove area classification remains challenging due to field inaccessibility and inefficiency of traditional assessment methods, highlighting the demand for advanced solutions. As the existing remote sensing-based studies suffer from limited classification accuracy and high computational complexity, this study combined Hyperspectral Image (HSI) with an Optimized Channel Level Residual CNN (OC-LRCNN) model for improved results in mangrove-related research. The proposed model employs unsupervised feature extraction to capture essential patterns with minimal training data while channel-level residual connections enhance discriminative feature selection and reduce spectral redundancy. Utilizing the Pichavaram EO-1 Hyperion and AVIRIS-NG datasets, the proposed model is compared with traditional CNN, state-of-the-art deep learning architectures (VGG, ResNet, DenseNet) and machine learning methods like SVM and RF. The OC-LRCNN achieved classification accuracies of 98.2% and 99.0% for the Hyperion and AVIRIS-NG datasets with consistently high precision, recall, F1-score and kappa values. These findings demonstrate the model’s effectiveness in reliable mangrove classification and monitoring applications.