Detecting stress caused by nitrogen deficit using deep learning techniques applied on plant electrophysiological data

Area of Science:

• Agricultural Science
• Plant Physiology
• Computational Biology

Background:

• Plant electrophysiology is a promising tool for assessing plant health.
• Traditional methods for analyzing electrophysiological data are computationally intensive and simplify raw signals.
• Deep learning (DL) offers automated feature learning but is underutilized for plant stress detection in electrophysiology.

Purpose of the Study:

• To apply deep learning techniques to raw electrophysiological data for detecting nitrogen deficiency stress in tomato plants.
• To evaluate the accuracy and efficiency of DL models compared to existing methods.
• To explore the potential for early stress detection in agricultural settings.

Main Methods:

• Utilized raw electrophysiological recordings from 16 tomato plants under typical production conditions.
• Applied deep learning models for automated classification of plant stress states.
• Assessed model performance using prediction accuracy and confidence levels.

Main Results:

• Achieved an initial prediction accuracy of approximately 88% for detecting plant stress.
• Increased accuracy to over 96% by incorporating prediction confidences.
• Demonstrated superior performance over the current state-of-the-art by over 8% accuracy.
• Successfully detected stress at its early stages.

Conclusions:

• Deep learning effectively analyzes raw plant electrophysiological data for stress detection.
• The proposed DL approach offers a significant improvement in accuracy and efficiency for identifying nitrogen deficiency.
• This method has direct applicability in production environments, paving the way for automated and sustainable agricultural practices.