Visible-NIR hyperspectral classification of grass based on multivariate smooth mapping and extreme active learning...

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Grassland management requires accurate community classification for animal husbandry and environmental monitoring.
• Visible-near infrared (NIR) hyperspectral imaging offers detailed spectral information but faces challenges with high costs and subjective labeling of samples.
• Existing methods struggle with the high dimensionality and detailed spectral information inherent in hyperspectral data.

Purpose of the Study:

• To develop an efficient and accurate visible-NIR hyperspectral classification model for grass communities.
• To address the limitations of expensive and subjective hyperspectral sample labeling.
• To enable intelligent and real-time grassland management through improved classification.

Main Methods:

• Proposed a multivariate smooth mapping (MSM) algorithm for spectral preprocessing and reconstruction, incorporating isometric feature mapping.
• Developed an extreme active learning (EAL) framework, combining XGBoost and active learning (AL), to intelligently select informative samples.
• Assembled a field hyperspectral collection platform to create the Grass1 dataset (nm resolution, 750 samples).

Main Results:

• The MSM preprocessing reduced processing time by 9.471 seconds compared to full spectrum analysis while maintaining overall accuracy (OA).
• The EAL framework significantly improved OA by 22.2% over traditional AL.
• The combined MSM-EAL model, specifically XAL, demonstrated superior performance in Kappa, Macro, Recall, and F1-score on the Grass1 dataset.

Conclusions:

• The lightweight MSM-EAL model provides an intelligent and real-time solution for high-precision grass classification.
• This approach offers a novel method for dynamic environmental monitoring and enhanced grassland management.
• The study validates the effectiveness of integrating advanced spectral preprocessing with active learning for hyperspectral data analysis.