A hyperspectral deep learning attention model for predicting lettuce chlorophyll content.

Area of Science:

• Plant physiology
• Agricultural science
• Computer vision

Background:

• Leaf phenotypic traits directly reflect agronomic traits crucial for high-quality leafy vegetable variety selection.
• Current research primarily focuses on morphological and structural plant traits, with limited investigation into physiological leaf phenotypes.
• Hyperspectral imaging offers potential for non-destructive physiological trait assessment.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting total chlorophyll content in greenhouse lettuce.
• To utilize full-spectrum hyperspectral images for chlorophyll estimation.
• To explore the application of spectral attention mechanisms in plant phenotyping.

Main Methods:

• A one-dimensional convolutional neural network (CNN) architecture was employed.
• A spectral attention module was integrated into the CNN model.
• The model was trained and validated using hyperspectral images of greenhouse lettuce.

Main Results:

• The developed deep learning model with a spectral attention module accurately estimated lettuce chlorophyll.
• The model achieved an average R² of 0.746 and an average RMSE of 2.018.
• Performance surpassed traditional methods like partial least squares regression (PLSR) and random forest (RF).

Conclusions:

• Deep attention networks combined with hyperspectral imaging provide a viable method for estimating lettuce chlorophyll.
• This approach enables convenient, non-destructive, and effective monitoring of leafy vegetable physiological traits.
• The findings support automatic monitoring and production management in leafy vegetable cultivation.