Clivia biosensor: Soil moisture identification based on electrophysiology signals with deep learning

Area of Science:

• Plant electrophysiology
• Environmental biosensing
• Machine learning for biological systems

Background:

• Plants generate electrical signals in response to environmental changes, indicating potential as biosensors.
• Challenges exist in correlating plant electrical signals with specific environmental data, particularly soil moisture gradients.
• Existing methods for signal classification are computationally intensive.

Purpose of the Study:

• To document and create a dataset of clivia electrical signals under varying soil moisture conditions.
• To develop and evaluate a lightweight convolutional neural network (CNN) for classifying plant electrical signals.
• To advance the development of plant-based biosensors for environmental monitoring.

Main Methods:

• Recorded electrical signals from clivia plants subjected to different soil moisture gradients.
• Compiled a dataset of these electrical signals for machine learning.
• Developed and trained a novel lightweight CNN model, named PlantNet, for signal classification.

Main Results:

• The PlantNet model achieved high classification accuracy (99.26%), precision (99.31%), recall (92.26%), and F1-score (99.21%).
• PlantNet demonstrated superior performance with significantly lower computational resource consumption (0.17M parameters, 7.17MB size, 14.66M FLOPs) compared to traditional CNNs.
• A robust dataset for plant electrical signal classification under soil moisture gradients was successfully created.

Conclusions:

• This research validates the potential of plants as effective biosensors for soil moisture detection.
• The developed lightweight CNN model offers an efficient solution for classifying plant-based biosensor signals.
• Findings provide a foundation for expanding plant biosensors to detect other environmental factors like pollutants (ozone, PM2.5, VOCs).