Scholarlink

<p>Accurate crop classification using hyperspectral data is essential for precision agriculture and environmental monitoring. This study presents a hybrid framework combining wavelet transform and parallel bidirectional long short-term memory (BiLSTM) networks. The wavelet transform is first employed to extract multi-resolution spatial features from the hyperspectral cube, which are then processed through parallel BiLSTM branches to capture forward and backward spectral dependencies. These outputs are combined and transmitted through a dense classification layer. The model was evaluated on five standard hyperspectral datasets; Indian Pines, Salinas, and Wuhan UAV borne hyperspectral image dataset series, which have three different datasets HanChuan, HongHu, and LongKou for crop classification, demonstrating high classification performance with overall accuracy of values ranging from 99.28%, 99.73%, 99.70%, 99.17% to 99.68%, respectively. The results underscore the effectiveness of combining spectral-spatial decomposition with sequential modeling for precise crop type identification.</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

Spectral-Spatial Crop Classification Using Wavelet-Based Parallel Bidirectional LSTM Networks

  • 期刊论文

  • 发表日期：

    2026年4月26日

查看全文
  • Ashish Kumar,
  • R. D. Garg

摘要

Accurate crop classification using hyperspectral data is essential for precision agriculture and environmental monitoring. This study presents a hybrid framework combining wavelet transform and parallel bidirectional long short-term memory (BiLSTM) networks. The wavelet transform is first employed to extract multi-resolution spatial features from the hyperspectral cube, which are then processed through parallel BiLSTM branches to capture forward and backward spectral dependencies. These outputs are combined and transmitted through a dense classification layer. The model was evaluated on five standard hyperspectral datasets; Indian Pines, Salinas, and Wuhan UAV borne hyperspectral image dataset series, which have three different datasets HanChuan, HongHu, and LongKou for crop classification, demonstrating high classification performance with overall accuracy of values ranging from 99.28%, 99.73%, 99.70%, 99.17% to 99.68%, respectively. The results underscore the effectiveness of combining spectral-spatial decomposition with sequential modeling for precise crop type identification.