Automated insect detection using acoustic features based on sound generated from insect activities
School of Engineering
This paper presents an automated insect detection technique using acoustic features and machine learning techniques based on sound signals generated from insect activities. The input sound signal was first pre-processed and segmented into windows frames from which the low-level set of signal properties and Mel-Frequency Cepstrum Coefficients were extracted. The detection accuracy of the features was tested on 11 insects of 6 species using a number of classifiers. The results have shown that a suitable acoustic feature set can be used to detect insects with high accuracy. Furthermore, the ensemble classifiers such as Bagged Tree provided the best accuracy in detecting both species classification (over 97.1%) and insect classification (over 92.3%). On the other hand, fine k-nearest neighbour classifier offered a balance between the quick training time (around 1 s) and the detection accuracy (over 88.5%).