Application Of Artificial Neural Network To Identify Species Of Mosquitoes (Diptera: Culicidae) Based On Wingbeat Frequency

Wingbeat waveforms of A. albopictus, A. aegypti, C. pipiens pallens, C. pipiens quinquefasciatus, and C. pipiens molestus were recorded by a photosensor (Qubit Systems, Inc., Canada, 2001) and a WfRer system (V2.0, WfRec. Qubit Systems, Canada, 2001). Spectrum characteristic of the wingbeat waveform was extracted signal-analysis software programmed in MATLAB. Artificial neural network classifiers were respectively built by wingbeat waveform time series, wingbeat frequencies and combination of them. Paper also researched effect on temperature and blood-fed on wingbeat frequency of the females, and analyzed correlations of weigh, length of wing and wingbeat frequency of the females.The result of record of wingbeat waveform showed that the wingbeat waveform of each species of the mosquitoes was analogical sine wave; Wingbeat frequencies were great difference and overlapped with different degree, whose arrangement was from 379Hz(female of C. pipiens quinquefasciatus) to 632Hz(male of C. pipiens pallens). Obvious diversity was found among different species of mosquitoes and between female and male of the same species with C. pipiens quinquefasciatus exception.There was evident diversity between wingbeat frequency of blood-fed female mosquito and non blood-fed one (df=58,p<0.001). The wingbeat frequencies of blood-fed female mosquitoes were average higher 49.61 Hz than non blood-fed one.The diversity of wingbeat frequencies of female mosquitoes was evident among five temperature levels (p<0.05). The result showed that there was effect on wingbeat frequencies of female mosquitoes. The result indicated that effect of high temperature was more evident than low temperature.The research of correlations among weighs, lengths of wing and wingbeat frequencies showed that there was no marked correlation between weigh and wingbeat frequencies (p>0.05), however, remarkable negative correlation between length of wing and wingbeat frequencies (p<0.01).The result of automated identification by artificial neural network showed that the most accurate classifiertested was an artificial neural network by using variable ofwingbeat frequency. Especially that was accurate for male of mosquitoes over 75% percent and relatively low accuracy for female of mosquitoes. The average accuracy was only 35.11% percent and 59.77% percent by artificial neural network that constructed by data of wingbeat waveform or wingbeat waveform and wingbeat frequency. So it was feasible to automated identification the species of mosquitoes by artificial neural network built by wingbeat frequency but data of wingbeat waveform not good. Hence, wingbeat frequencies were significant for different species of mosquitoes to recognize each other.