Hyperspectral inversion model of ginkgo leaf yield prediction based on machine learning.

Area of Science:

• Agricultural Science
• Remote Sensing
• Plant Physiology

Background:

• Ginkgo biloba leaf yield is crucial for assessing plant health and growth.
• Traditional manual yield assessment methods are inefficient, labor-intensive, and costly.

Purpose of the Study:

• To develop an accurate, non-destructive, and efficient method for estimating ginkgo leaf yield using airborne hyperspectral imaging.
• To optimize hyperspectral data preprocessing and feature selection for yield prediction.

Main Methods:

• Airborne hyperspectral imaging was used to collect canopy spectral data.
• Various preprocessing techniques (MSC, SNV, SG, FD, SS) and feature selection algorithms (PSO, SPA, PCA, LASSO, CARS, PSAMA) were evaluated.
• Machine learning models including PLSR, RF, KNNR, LSTM, SVR, BiLSTM, and BiLSTM-GS were employed for yield prediction.

Main Results:

• The SNV-PLSR model showed good performance (Rp² = 0.7831).
• An optimized model combining SNV preprocessing, selected vegetation indices (SAVI, MSAVI, NDRE, SIPI), Region of Interest Pixels (ROP), and the BiLSTM-GS algorithm achieved high prediction accuracy (Rp² = 0.8795).

Conclusions:

• Airborne hyperspectral canopy-based estimation is a viable and accurate technology for monitoring ginkgo leaf yield.
• This non-destructive approach significantly improves upon traditional manual methods.