A Comprehensive Survey on Bird Species Identification Models

Abstract

Fundamental classes of ecosystem comprise of provisioning, regulating, supporting, and cultural services. Of all the living organisms, ‘birds’ adhere to all the earlier mentioned classes thereby motivating the current work in this paper. Humans coexist with birds and generally identify them via either their physical appearance or their vocal note. Bird species identification (BSI) has attracted global attention lately owing to the rise of technological prowess in aiding their identification. Although birds are identified by their physical appearance and vocal notes; often these two features lead to ambiguity due to image acquisition in a diverse range of postures, intensity, and varied notes of the vocal data. Recent models trained on Deep Neural Networks (DNN) have exhibited better performance in minimizing the said ambiguity. In-depth analysis of proposals in BSI mandates focused study of such prior art, majorly, to identify the inaccuracies generated through ambiguity. The present work provides the reader a detailed survey of preceding art addressing both the dimensions of BSI, namely physical appearance and vocal note based. The analysis stresses mostly on the DNN based models focused on Image Isolation, Background Reduction for physical appearance identification, and all those proposals which evaluate the vocal note datasets. The work in this paper targets those researchers who intend to pursue further study in BSI and design more efficient models and serves as a benchmark for future exploration in BSI using physical appearances and vocal notes. This work provides adequate information for amateur researchers to explore the research on bird species identification models.