Sensor identifies insects by wingbeat frequency

A new device more rapidly, precisely and inexpensively identifies mosquitoes that transmit dengue fever and yellow fever, as well as insects that cause agricultural infestations.

By Elton Alisson

Agência FAPESP – Health surveillance services in Brazil and other countries may soon be aided by a technology that works more rapidly, precisely and inexpensively to identify foci of mosquitoes that carry diseases such as dengue fever, malaria and yellow fever.

A group of researchers from the Computational Intelligence Laboratory of the Institute of Mathematical Sciences and Computation (ICMC) at the University of São Paulo (USP), São Carlos campus, together with colleagues from Bourns College of Engineering at the University of California Riverside (UCR) and the US subsidiary of the Brazilian company Isca Tecnologias, has developed a sensor that can automatically identify and quantify several species of flying insects that cause diseases or agricultural infestations.

The result of a project carried out with funding received from FAPESP, the Bill & Melinda Gates Foundation and the Vodafone Americas Foundation, the sensor was described in an article published in the June issue of the Journal of Insect Behavior.

“The sensor allows monitoring populations of insects that are harmful to the health of humans or damage agriculture and the environment, in a faster, smarter and more precise way,” Gustavo Enrique de Almeida Prado Alves Batista, professor at the ICMC and project coordinator, told Agência FAPESP.

“Instead of spraying insecticide over an entire region where a particular species of harmful flying insect or its larvae may be, it can be applied just to areas identified by the sensor as insect foci,” he explained.

Development of the apparatus began in 2010, when Batista started his post-doctoral work at the UCR with a scholarship from FAPESP, in collaboration with a group led by Eamonn John Keogh, professor of computer sciences at the UCR, along with Agenor Mafra-Neto, principal investigator at Isca Tecnologias.

At the time, Keogh was interested in developing a system to automatically classify insects based on voice recognition and machine learning techniques – an area of artificial intelligence dedicated to the development of algorithms (sequences of commands) and techniques that enable a computer to improve its performance in executing tasks.

The solution developed by Batista in partnership with Keogh’s group was a laser sensor based on analysis of the acoustic frequency of insects’ wingbeats during flight.

“Flying insects beat their wings at different speeds, according to their size and other morphological characteristics, and at acoustic frequencies that typically vary between 100 and 1,500 Hertz,” Batista explained. “Our idea was to develop a system that could identify the acoustic frequency of various flying insect wingbeats in addition to other information we could use to classify the insects.”

Sensor function

The researchers developed a sensor that consists of a low-powered laser beam aimed at an array of phototransistors – similar to a laser beam pointed at a wall.

As the insect flies between the laser beam and the phototransistor array, its wings partially block the light and cause small variations in it.

Oscillations in the light caused by an insect’s wings are captured by the phototransistor array. The signals are similar to audio signals, such as those captured by a conventional microphone. The difference is that they are produced not by the variation of the sound waves but by the variation of the light.

The signals extracted by the sensor are filtered and amplified through an electronic circuit board. Using a digital sound recorder connected to the board output, it is possible to record the signals in audio files and transfer them to a computer for analysis.

“Each species of flying insect produces a slightly different signal. This allows a computational comparison of each of the different species,” Batista said.

The data for calibrating and classifying species through the sensor were collected by placing the insects in acrylic boxes containing mounted sensors, with controlled lighting, temperature and humidity.

Each box with a sensor received dozens of pre-classified flying insects belonging to a single species. These included the mosquitoes Aedes aegypti (which transmits dengue and yellow fever), Anopheles gambiae (malaria vector), Culex quinquefasciatus (lymphatic filariasis vector) and Culex tarsalis (Saint Louis encephalitis and western equine encephalitis vector), in addition to the fly Drosophila melanogaster (known by its common name “banana fly”), the fly Musca domestica, certain insects belonging to the family Psychodidae (known as drain flies) , the beetle Cotinis mutabilis and the bee species Apis mellifera.

After collecting data for 15 days, the researchers recorded the signals generated by the simple passing of the insects through the sensor’s laser beam inside the acrylic boxes, deleting any background noise. The signals obtained by the sensors in the various boxes of insects were then mixed and recorded into a single file.

When the audio file was analyzed by software that included a classification algorithm, also developed by the researchers, the computer system was able to differentiate and identify the species of insect with 98–99% accuracy.

“Now, we’re only exploring the frequency of wingbeats and other attributes intrinsic to the sensor signal,” Batista said. “There are other variables that may be added to better understand the sensor’s success rate in identifying species of insects.” These variables include the time of day at which the insects fly in addition to air temperature, pressure and humidity – the three meteorological factors that most affect insect activity.

It is thought that an increase in temperature causes changes in metabolism and increases the wingbeat frequency of the insects, Batista noted.

Through a study carried out by post-doctoral candidate Vinícius Mourão Alves de Souza, another FAPESP scholarship recipient, the researchers are studying how the signal obtained by the sensor varies according to the environmental conditions the insects are in. Batista said, “We want to analyze how the sensor functions under differing conditions of temperature, humidity and air pressure.”

In a study conducted by Diego Furtado Silva, another FAPESP scholarship recipient, the researchers extracted additional data (on attributes) from the signals that might provide information beyond wingbeat frequency. “We’re using a series of techniques based mainly on voice recognition to extract better attributes than just wingbeat frequency,” Batista explained.

Intelligent trap

The laser sensor was used in a prototype intelligent trap designed by ICMC researchers in collaboration with the subsidiary of Isca Tecnologias in Riverside.

The device is able to identify flying insects in real time, using a laser sensor to capture target species, includinginsects that transmit diseases and agricultural pests. The device also allows non-harmful insects, such as bees and other pollinators or food sources for other animals, to be released back into the environment.

“Since we began developing the sensor, we had the idea of using it in practical applications, such as an intelligent insect trap,” Batista said.

The trap is cylindrical and made of ABS pipe with a laser sensor mounted at its entry point and a collection bag at its exit point – much like a vacuum cleaner. It has a valve at the entry point that releases carbon dioxide – a substance capable of attracting the females of many species of mosquitoes.

When flying in front of the trap’s entry point, the insect is sucked in by an airflow generated by a fan similar to that found in a computer, which carries it to a chamber containing the laser sensor for classification.

If the insect is identified as a non-harmful species, an exit door opens and the insect is pushed out of the trap by an inversion of the airflow.

However, if it is identified as a harmful species, the airflow pushes the insect into the collection bag, where it is held by adhesive paper similar to that used in conventional non-selective adhesive traps – those that capture all species of insects, including harmless ones.

“The trap enables easier and more precise identification and quantification of undesirable insects in a particular area,” Batista said. “This is how it’s possible to monitor the harmful insect population in a particular region in real time and report this information through wireless networks to health-surveillance agencies.”

Low cost

The researchers estimate that the sensor has potential for broad use due to its low manufacturing cost – less than R$30 – and the fact that it is powered by solar energy or even a battery.

In the health sector, one of the device’s main applications may be fighting Anopheles mosquitoes, which are vectors of malaria, and Aedes mosquitoes, which transmit dengue fever and yellow fever.

The article “Flying insect classification with inexpensive sensors” (doi: 10.1007/s10905-014-9454-4), by Batista and colleagues, can be read by subscribers to the Journal of Insect Behavior at http://link.springer.com/article/10.1007/s10905-014-9454-4.