Self-supervised Animal Detection in Indoor Environment

Taifour Yousra, Hina Afridi, Adane Nega Tarekegn, Mohib Ullah, Azeddine Beghdadi, Faouzi Alaya Cheikh

Abstract: Animal detection holds great potential for applications in animal farming, conservation and ecological studies. However, it faces challenges such as limited data availability and varying environmental conditions. Then challenges pose difficulties in developing precise detection models using traditional supervised learning techniques that rely on large amounts of labelled data. Moreover, factors like illumination changes and occlusion due to animal motion further complicate the detection process. This study explored a novel approach based on self-supervised learning for animal detection. In a nutshell, we leveraged various self-supervised learning methods to extract meaningful features from unlabeled data, which were then utilized for animal detection using our custom-built dataset. Our research sheds light on the efficacy of self-supervised learning in tackling the difficulties posed by a scarcity of labelled data and environmental fluctuations. Moreover, our methodology provides valuable insights for enhancing animal detection in studies related to industrial farming.