Mosquito causes spread of several contagious diseases worldwide such as Malaria, Dengue, Zika, Chikungunya and yellow fever. Dengue is a very challenging disease to control the speared and it causes several deaths every year. One of the methods to control the spread is by controlling the spread of mosquitos, specifically the type of mosquito that carries Dengue virus. It is already identified that Dengue virus is spread in Sri Lanka by the type of mosquito called Aedes species. So one way of controlling the disease is by correctly identifying the spread of Aedes species. For the identification of different types of mosquitos mostly image based visual methods are used. Very few methods use acoustic features such as fundamental wing-beat frequency data. This proposed work intends to automatically classify Dengue mosquito and non-Dengue mosquito using the wing-beat sound of mosquitos and it targets to develop an automatic system to classify the above two classes and ultimately to develop a mobile app that could make a warning sound whenever there is a Dengue mosquito in the acoustic vicinity of the phone.

For this automatic classification of Dengue against non-Dengue mosquitos initially MFCC (Mel Frequency Cepstral Coefficients) feature which is a popular feature in speech based automatic systems is used. As the back-end classifier, (DNN) deep neural network is used. This system produced a results of 75% of accuracy. Targeting to derive a more specific feature for this classification the spectral contents of the mosquito sounds are compared. For this comparison a set of filter bank is used to extract normalized energy in each band from all the mosquito sounds in the dataset. Then this normalized energy is averaged across all sound recording for Dengue and Non-Dengue separately. This normalized energy is shown in Figure 1. From the figure it can be observed that two distinct spectral area can be seen as more discriminative. As the future work the MFCC feature is planned to be modified by allocating more filters in those discriminative regions and having less filters in other area. 

Figure 1: Normalized energy plot of dengue and non-dengue mosquitoes

Keywords: Aedes, MFCC, DNN 

References

[1] H. Mukundarajan, F. J. H. Hol, E. A. Castillo, C. Newby, and M. Prakash, “Using mobile phones as acoustic sensors for high-throughput mosquito surveillance,” Elife, vol. 6, pp. 1–26, 2017, doi: 10.7554/eLife.27854.

[2] Y. Li et al., “Mosquito detection with low-cost smartphones: data acquisition for malaria research,” 2017, [Online]. Available: http://arxiv.org/abs/1711.06346.

[3] A. Lukman, A. Harjoko, and C. K. Yang, “Classification MFCC feature from culex and aedes aegypti mosquitoes noise using support vector machine,” Proc. - 2017 Int. Conf. Soft Comput. Intell. Syst. Inf. Technol. Build. Intell. Through IOT Big Data, ICSIIT 2017, vol. 2018-Janua, pp. 17–20, 2017, doi: 10.1109/ICSIIT.2017.28.