Detecting and Classifying Hawaiian Monk Seal Vocalizations using Deep Learning
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3020384Utgivelsesdato
2022Metadata
Vis full innførselSamlinger
Beskrivelse
Full text not available
Sammendrag
The Hawaiian monk seal (Neomonachus schauinslandi) is one of the most endangered ma-rine mammals in the world. With an estimated population of 1,570 individuals remainingin the wild, the species have an urgent conservation priority. For effective management ofthreatened and endangered species, monitoring can prove crucial for detecting possible unex-pected changing in their environment. The majority of the Hawaiian monk seal’s populationexist in the marine conservation area located in the Northwest Hawaiian Islands. Therefore,non-invasive methods, such as passive acoustic monitoring, could be a possible solution tomonitoring the species in their habitat.However, analyzing the vast quantities of data obtained with passive acoustic monitoring isslow and labor-intensive. In this thesis, we present a deep learning-based approach to auto-mate the process. With the use of a Region-Based Convolutional Neural Network (R-CNN)to detect the presence of a Hawaiian monk seal’s underwater vocalization. Followed by aConvolutional Neural Network to classify the vocalizations into six previously defined calltypes. The detector managed to achieve an AP of 0.35 and AP50 of 0.55, meanwhile theclassifier achieved an accuracy of 93.3%