Plant Bioelectrical Signals for Environmental and Emotional State Classification.

Area of Science:

• Plant bioelectricity
• Human-computer interaction
• Machine learning for biosignal analysis

Background:

• Plants generate bioelectrical signals that can be influenced by their environment and potentially by external stimuli.
• Previous research has explored plant responses to environmental changes, but less is known about their potential for complex classification tasks like emotion recognition.

Purpose of the Study:

• To investigate the feasibility of using bioelectrical signals from a single Purple Heart plant (Tradescantia pallida) for dual-purpose classification.
• To assess the plant's ability to detect environmental states (e.g., lamp on/off) and recognize human emotional states (happy/sad).

Main Methods:

• Bioelectrical signals were recorded from a Tradescantia pallida using an AD8232 ECG sensor at a 400 Hz sampling rate.
• Signal segments were converted into mel-spectrograms and classified using a ResNet18 Convolutional Neural Network (CNN).
• The system was tested on environmental state detection (lamp on/off) and human emotion classification (happy/sad).

Main Results:

• Environmental state detection (lamp on/off) achieved 85.4% accuracy with balanced precision and recall.
• Human emotion classification achieved 73% accuracy with a 1-second lag, distinguishing between happy and sad states.
• The study demonstrated preliminary feasibility but highlighted limitations including a single-plant setup and reliance on external labeling.

Conclusions:

• The pilot study suggests that plant bioelectrical signals hold potential for environmental sensing and rudimentary human emotion recognition.
• Further research with larger sample sizes, multiple plants, and diverse experimental conditions is necessary to validate these findings and improve generalizability.
• This work opens avenues for exploring novel human-plant communication interfaces and biosensing applications.