Acoustic Classification of Insects using Signal Processing and Deep Learning Approaches

Abstract

Classification of insects is an essential aspect of Entomology, the study of insects. It also has several applications in agriculture, where insects play a crucial role in both positive and negative ways. This paper presents a signal processing and deep learning approach for insect classification based on the sounds produced by the insects. In the signal processing approach, features are extracted based on a Mel-spectrogram using an FFT window of length 2048, hop length of 512, and 128 mel bands. The 2D-CNN is used further to classify the Mel-spectrograms into their respective insect classes. Further, instead of extracting hand-crafted features manually, automatic feature extraction and classification were done through a deep learning approach. Two deep-learning models, 1D-CNN and SincNet, have been employed to classify the insects based on their acoustic sounds. A comparative performance analysis between handcrafted and deep learning features reveals that SincNet based deep learning model produced the best accuracy of 97% on the publicly available Cornell University Insect database.