Leafhoppers are among the chief transmitters of plant disease. In the species,
the recognition and discovery of potential partners takes place exclusively through
species- and sex-specific vibrational signals taking place in precisely coordinated
duets. This work describes an autonomous system AS, capable of recognizing the
male call of the leafhopper Aphrodes bicincta “Dragonja” and generating female
replies in real time. The species in question was selected as it has a complex duet
structure, with the female replies having to appear in short (4–175ms) intervals
between continuously repeated elements in the male call, triggering the male
searching behaviour. As the male call is transmitted via the plant, a variation in
the frequency parameters of the registered call can be observed during the search.
Designed on the basis of a digital signal controller was special hardware, with
corresponding software for the AS. The AS algorithm is based on human speech
recognition methods comprising of feature extraction and classification. The
features used were linear prediction cepstral coefficients, chosen after a computational
time comparison with the linear frequency cepstral coefficients. The latter
method was selected as the basis of the behavioural experiment based on a simulation
of three different classifiers: the Gaussian mixture model, support vector
machines and multilayer perceptron. An algorithm for preparing the training set
for supervised classifier learning was devised. A bandwidth-limited linear prediction
call activity detector based on spectrum peak tracking was used to prevent
feeding the noise-based feature vectors into the classifier. For faster analysis of
the audio recordings from the behavioural experiment, an algorithm was devised
to automatically extract the relevant characteristic parameters. The described
algorithm can also be used on other species with a similar frequency - temporal
vibrational signal structure.
We tested the efficiency of the AS in behavioural experiments with live males.
The AS method successfully classified vibrational calls of a male A. bicincta
“Dragonja” from the background noise. There was no statistical difference in
percentage of males that established a duet between the AS and the live female.
The mimicking of a duetting female by the AS also attracted the males to the
source of the female reply. Simulations of the spectral subtraction method and
the computational time measurements on the AS were based on the behavioural
experiment results. The described method improved the signal-to-noise ratio of
the male call recordings by 26 dB and was deemed appropriate for real-time signal
recognition on the AS.
Also researched were plant equalization methods based on the inverse playback
signal filtering. By comparing the simulations of the LMS adaptive filtering
method and the linear prediction method, the latter was chosen for tests on the
plant. The linear prediction method successfully reduced the influence of the plant
on the frequency parameters of the playback signal at the point of measurement.
The primary purpose of the AS is to further study vibrational communication.
Based on the study results, different pest control systems could be developed
which would use vibrational communication instead of pesticides. Despite current
limitations of the vibrational signal measurement and reproduction, the AS
represents a starting point in the development of a vibrational trap which would
attract and capture males and disrupt the reproduction cycle of pests.
|
