Izvorni znanstveni članak

https://doi.org/10.17559/TV-20250816002903

A Fuzzy-Driven Deep Learning Framework with Spectral-Spatial Attention for Efficient Hyperspectral Data Classification in Precision Agriculture

Barath S. ; School of Computer Science, Takshashila University, Tindivanam Taluk, Ongur, Tamil Nadu 604305, India *
Senthil M. ; School of Computational Engineering, Takshashila University, Tindivanam Taluk, Ongur, Tamil Nadu 604305, India

* Dopisni autor.

Sažetak

Hyperspectral sensing provides highly detailed information on crops, soil, and environmental conditions by capturing reflectance data across numerous spectral bands. This capability enables precise monitoring and decision-making in agriculture but also introduces challenges due to the high dimensionality and complexity of hyperspectral data. To address these issues, this study proposes a novel Fuzzy-Driven Adaptive Deep Convolutional Network (FADCNet) for accurate hyperspectral image classification. The framework integrates multiple innovative components to enhance both efficiency and accuracy. First, a Spectral-Spatial Attention Mechanism (SSAM) is employed to emphasize discriminative spectral and spatial features, ensuring that important patterns in wavelength and location are preserved. Second, an Entropy-Guided Band Selection (EGBS) method is introduced to eliminate redundant and noisy bands using information-theoretic criteria, thereby improving computational efficiency without compromising classification quality. Third, an Adaptive Stochastic Gradient with Momentum Optimizer (ASGMO) is designed to accelerate learning by dynamically adjusting learning rates and incorporating momentum, which stabilizes training and mitigates convergence issues. The proposed framework is tested on a real-time hyperspectral dataset collected from agricultural regions in Tamil Nadu, India. Preprocessing steps, including noise suppression and normalization, are applied to enhance data quality before analysis. Comparative experiments with state-of-the-art techniques highlight that FADCNet consistently achieves superior performance in terms of classification accuracy, precision, recall, F1-score, and processing efficiency. These results confirm the effectiveness of the proposed framework for agricultural monitoring and suggest its broader applicability in hyperspectral image analysis across various domains.

Ključne riječi

deep learning optimization; entropy-guided band selection; hyperspectral data classification; precision agriculture; spectral-spatial attention mechanism;

Hrčak ID:

353067

URI

https://hrcak.srce.hr/353067

Datum izdavanja:

7.11.2026.

Posjeta: 0 *