Stressed plants emit ultrasonic noises that can be identified with ML

According to a recent study, plants emit ultrasonic sounds in response to specific stress factors like drought or cutting. The researchers from Tel Aviv University have used the emitted sounds to develop machine learning models that could identify the plants’ condition, such as dehydration level and injury.

The study provides novel insights into the interactions between plants and their environment, and may also have practical implications for plant monitoring in agriculture.

Image generated by AI

Plants produce visual, chemical, and tactile cues that function as signals to their environment and are detectable by other organisms.

The researchers in this study sought to explore the possibility of plants emitting informative airborne sounds that could also serve as signals or cues to their surroundings.

The approach involved training machine learning models to classify different plant species and their conditions based on the ultrasonic sound emissions.

The sounds were divided into four groups based on two plant types (tomato or tobacco) and two treatments (drought stress or cutting).

The results show that stressed plants emit informative airborne sounds that can be detected from a distance of several meters, in both acoustic chambers and greenhouses. Moreover, the sounds carry valuable information about the plant’s physiological state.

The plants, mostly tomato and tobacco, were recorded when cut and dehydrated

Method and results

The research was conducted in two stages:

First stage: generating a classifier

To perform the acoustic tests, the researchers placed 3 plants in a custom-made wooden box that was equipped with acoustic foam tiles for the purpose of capturing the sounds generated by the plants placed inside it. The box was located in a quiet basement and designed to minimize echoes. Three pairs of microphones were placed inside the box, directed at each plant stem, from a distance of 10 cm.

They then trained a convolutional neural network (CNN) classifier to distinguish between the tomatoes sounds recorded in the acoustic box and the empty greenhouse noises. The model achieved 99.7% balanced accuracy in a cross-validation examination.

Plants tested in an “acoustically isolated box” without background noise

Second stage: greenhouse experiment

The researchers recorded sounds from tomato plants in the greenhouse and then used the trained CNN classifier to filter the sounds. The model achieved a balanced accuracy score of 99.7% in correctly distinguishing between tomato sounds and background noises.

Even in a noisy greenhouse filled with sounds of people talking and building renovations next door, the algorithm was able to distinguish between different plants.

Acoustic manifestation of dehydration in tomato plants recorded in a greenhouse

Conclusion, further research

This study demonstrates that stressed plants emit informative sounds that can potentially be used to identify the condition of the plants, such as their dehydration level and injury, and may also be detectable by other organisms.

Although there is no evidence to suggest that plants actively communicate with each other through sound,  it is possible that these sounds may be detectable by other organisms.

This research can pave the way for a better understanding of how plants interact with their environment, and how this knowledge can be used to improve agricultural practices.

Upcoming research could potentially devise novel techniques to monitor and maintain plant heath, ultimately resulting in enhanced crop production and improved food security.

Learn more:

Other popular posts