Automated hyperspectral vegetation index derivation using a hyperparameter optimisation framework for high-throughput...

Area of Science:

• Agricultural remote sensing
• Plant science
• Spectroscopy

Background:

• Hyperspectral vegetation indices (VIs) are crucial for estimating plant traits in agriculture and phenotyping.
• Current VIs, often simple two-band indices, have limited performance and generalizability for diverse traits.

Purpose of the Study:

• To develop an automated system (AutoVI) for generating novel, trait-specific hyperspectral VIs.
• To streamline the creation, optimization, and evaluation of VIs using the Tree Parzen Estimator algorithm.

Main Methods:

• The AutoVI system was employed to generate novel two- to six-band indices.
• The generated indices were tested for estimating chlorophyll and sugar content in wheat.
• Indices were used in multiple linear regressions and compared against existing VIs and partial least squares regression.

Main Results:

• AutoVI successfully generated complex, novel VIs (≥4 bands) with strong correlations (R² > 0.8) to wheat chlorophyll and sugar content.
• The AutoVI-derived indices significantly outperformed 47 existing VIs and partial least squares regression in estimation accuracy.
• The system demonstrated rapid generation of trait-specific indices.

Conclusions:

• The AutoVI system provides a powerful tool for developing advanced, trait-specific VIs for plant phenotyping.
• These novel VIs are readily adoptable for high-throughput phenotyping platforms, benefiting plant scientists and breeders.
• AutoVI enhances the capability of hyperspectral remote sensing in agriculture.