Classification Efficiency of Pre-Trained Deep CNN Models on Camera Trap Images

Adam Stančić; Vedran Vyroubal; Vedran Slijepčević

doi:10.3390/jimaging8020020

Classification Efficiency of Pre-Trained Deep CNN Models on Camera Trap Images

J Imaging. 2022 Jan 20;8(2):20. doi: 10.3390/jimaging8020020.

Authors

Adam Stančić¹, Vedran Vyroubal¹, Vedran Slijepčević²

Affiliations

¹ Department of Engineering, Karlovac University of Applied Sciences, Ivana Meštrovića 10, 47000 Karlovac, Croatia.
² Department of Wildife Management and Nature Protection, Karlovac University of Applied Sciences, Trg J. J. Strossmayera 9, 47000 Karlovac, Croatia.

Abstract

This paper presents the evaluation of 36 convolutional neural network (CNN) models, which were trained on the same dataset (ImageNet). The aim of this research was to evaluate the performance of pre-trained models on the binary classification of images in a "real-world" application. The classification of wildlife images was the use case, in particular, those of the Eurasian lynx (lat. "Lynx lynx"), which were collected by camera traps in various locations in Croatia. The collected images varied greatly in terms of image quality, while the dataset itself was highly imbalanced in terms of the percentage of images that depicted lynxes.

Keywords: CNN; camera trap; classification; efficiency; pre-trained.

Grants and funding

LIFE16 NAT/SI/000634/LIFE Lynx