Insect identification based on bioacoustic signal using spectral and temporal features

This paper presents an automatic identification algorithm for insect species by their acoustic signals. Insect identification is a primary need of entomologists both for research applications and for pest control. Cicadas are one of the most groups of singing insects. The species-specific songs of cicadas play a vital role in the species recognition. Acoustic characteristic is useful in the identification of cicadas, particularly when other characters are unavailable due to capture problems of cicadas. In this study, five species of cicadas are identified by their sounds. The acoustic signal of each cicada has been characterized by twenty-nine features in both spectral and temporal domains. To optimize the features, the genetic algorithm and outlier omitting method are used and the selected features are introduced to multi-layer perception classifier. The results of insect identification algorithm show the superiority of the proposed algorithm with the accuracy rate of 99.13%. Thus, it clearly demonstrates the utility and effectiveness of acoustic for identifying the cicada species in the agriculture applications.