Article Number: DRJAFS72902107
DOI: https://doi.org/10.26765/DRJAFS72902107
ISSN: 2354-4147

Vol. 10(11) Pp. 254-261, November 2022
Copyright © 2022
Author(s) retain the copyright of this article
This article is published under the terms of the

Creative Commons Attribution License 4.0.

Full-Text (PDF)


Original Research Article

Towards the Conservation of Endangered Mammals using Single-stage Deep Neural Network

Ejiyi, Chukwuebuka Joseph

Orakwue, Chiduzie O.

Qin, Zhen*

Diokpo, Chidinma N.

Nnani, Ann O.

Ejiyi, Makuachukwu B.

Goshu, Hana L.

Okpara, Chidimma P.


Abstract

For the purpose of conserving and preserving endangered mammals in their habitat, their real-time detection is very crucial. This will help curb their involvement in accidents as well as help in tracking them in case they stray which consequently will be useful for the preservation of life and property as most of the mammals are known to destroy crops among other damages that can be caused by them. We deployed a single-stage deep neural network – You Only Look Once (YOLO) version 4 for the detection of four classes of mammals in real-time which will help in monitoring the animals as well as tracking their activities not excluding their security and protection from poisoning and being preyed upon. A system of this magnitude can be employed to replace the manual monitoring of these mammals and the efforts put in place to track their position which can be achieved using a video in this model. The choice of YOLO has been because of its applicability for the detection of the 80 classes of objects in the MS COCO dataset and the high speed it has in performing this task. To be used for our purpose, we carefully choose the components of the network and modified and retrained the model to suit our desire which is to detect 4 classes of mammals that we have in our dataset after they were annotated. This research presents a custom dataset for the detection of this class of animals in real-time with a laudable mean Average precision of 98.6% @ IoU = 0.5 and an average real-time speed of detection of 30.6 FPS on a GPU-enabled computer and over 20FPS for a computer that is only CPU-enabled. These results show that the model produced is applicable in real-time for detection both in GPU- and CPU-enabled devices. Paper code url: https://github.com/GREAT0/4–Mammals-Detection.git

Full-Text (PDF)


Keywords: Conservation; Deep neural network; Mammals; Real-time; YOLO


 Received: October 1, 2022  Accepted: October 29, 2022  Published: November 1, 2022


Copyright © 2022 Direct Research Journal of Agriculture and Food Science