An approach towards plant electrical signal based external stimuli monitoring system

Abstract

Plants have sensing mechanisms which are employed to monitor their environment for optimal growth. This sensing mechanism can be observed by the change in behaviour in plants like Mimosa pudica (Touch Me Not) which closes its leaves when touched or Dionaea muscipula (Venus Flytrap) which closes its trap when an insect gets in it. It has been established that plants produce an electrical signal response to stimuli that is used to control various physiological phenomena within the plant. If such electrical signals are extracted and analysed, information about the external stimuli which caused the electrical signal may be found. If such an analysis is successful, then plants can be used as a living multiple stimuli sensor.
This work explores the possibility of extracting information from the plant electrical signal response to the external stimuli which caused the plant to produce such a signal. Initially, the plant was treated as a black box system and a simple input (light pulse as stimulus) – output (electrical signal response) system was modelled through system identification techniques. Thereafter, an inverse system was modelled for input (electrical signal response) – output (light pulse as stimulus) to find out if there exists, within the plant’s electrical signals, adequate information about the time of application and the intensity of the applied stimulus.
Next, classification methods were employed to find out if there was adequate information, within the raw plant electrical signal response, about the type of stimulus applied to the plants. More complex stimuli such as Sulphuric acid, Ozone and Sodium chloride solutions were applied to the plants to find out if the plant electrical signal response could be used to classify these stimuli in a binary classification scenario. Discriminant analysis based classifiers were employed along with simple statistical features which produced classification accuracy of around 70%.
A decision tree based classification strategy was then explored, using discriminant analysis lassifiers and statistical features, in a multiclass classification strategy with the aim of enhancing classification accuracy. This exploration involved more datasets which enabled a prospective study (separate data held out) to be carried out to see the results in a more realistic scenario. The decision tree based classification system produced an accuracy of around 90% for both retrospective and prospective studies. In this work, both raw and filtered signals were used, of which the raw signals produced marginally better results than the filtered ones.
Lastly, curve fitting coefficients were explored for classification of stimuli by fitting four different curves to raw plant electrical signals. Classification accuracy of around 90% was achieved during the retrospective study by using polynomial curve fit coefficients. This enabled features to be extracted from the entire duration of the time series rather than small segments of it, in order to see if classification accuracy improved.

More information

Identifiers

Catalogue record

Export record