This article is cited in 3 papers

IMAGE PROCESSING, PATTERN RECOGNITION

Classification of plumage images for identifying bird species

A. V. Bel'ko^a, K. S. Dobratulin^ab, A. V. Kuznetsov<sup>ac

a</sup> Samara National Research University
^b National University of Science and Technology «MISIS», Moscow
^c Image Processing Systems Institute of the RAS - Branch of the FSRC "Crystallography and Photonics" RAS, Samara, Russia, Samara

Abstract: This paper studies the possibility of using neural networks to classify plumage images in order to identify bird species. Taxonomic identification of bird plumage is widely used in aviation ornithology to analyze collisions with aircraft and develop methods for their prevention. This article provides a method for bird species identification based on a dataset made up in the previous research. A method for identifying birds from real-world images based on YoloV4 neural networks and DenseNet models is proposed. We present results of the feather classification task. We selected several deep learning archi-tectures (DenseNet based) for a comparison of categorical crossentropy values on the provided dataset. The experimental evaluation has shown that the proposed method allows determining the bird species from a photo of an individual feather with an accuracy of up to 81.03 % for accurate classification, and with an accuracy of 97.09 % for the first five predictions.

Keywords: machine vision, pattern recognition, neural networks, aviation ornithology

Received: 14.11.2020
Accepted: 02.07.2021

DOI: 10.18287/2412-6179-CO-836