Improving the precision and accuracy of animal population estimates with aerial image object detection

Jasper A.J. Eikelboom*, Johan Wind, Eline van de Ven, Lekishon M. Kenana, Bradley Schroder, Henrik J. de Knegt, Frank van Langevelde, Herbert H.T. Prins

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

Abstract

Animal population sizes are often estimated using aerial sample counts by human observers, both for wildlife and livestock. The associated methods of counting remained more or less the same since the 1970s, but suffer from low precision and low accuracy of population estimates. Aerial counts using cost-efficient Unmanned Aerial Vehicles or microlight aircrafts with cameras and an automated animal detection algorithm can potentially improve this precision and accuracy. Therefore, we evaluated the performance of the multi-class convolutional neural network RetinaNet in detecting elephants, giraffes and zebras in aerial images from two Kenyan animal counts. The algorithm detected 95% of the number of elephants, 91% of giraffes and 90% of zebras that were found by four layers of human annotation, of which it correctly detected an extra 2.8% of elephants, 3.8% giraffes and 4.0% zebras that were missed by all humans, while detecting only 1.6 to 5.0 false positives per true positive. Furthermore, the animal detections by the algorithm were less sensitive to the sighting distance than humans were. With such a high recall and precision, we posit it is feasible to replace manual aerial animal count methods (from images and/or directly) by only the manual identification of image bounding boxes selected by the algorithm and then use a correction factor equal to the inverse of the undercounting bias in the calculation of the population estimates. This correction factor causes the standard error of the population estimate to increase slightly compared to a manual method, but this increase can be compensated for when the sampling effort would increase by 23%. However, an increase in sampling effort of 160% to 1,050% can be attained with the same expenses for equipment and personnel using our proposed semi-automatic method compared to a manual method. Therefore, we conclude that our proposed aerial count method will improve the accuracy of population estimates and will decrease the standard error of population estimates by 31% to 67%. Most importantly, this animal detection algorithm has the potential to outperform humans in detecting animals from the air when supplied with images taken at a fixed rate.

Original languageEnglish
Pages (from-to)1875-1887
JournalMethods in Ecology and Evolution
Volume10
Issue number11
Early online date2 Aug 2019
DOIs
Publication statusPublished - Nov 2019

Fingerprint

Giraffa camelopardalis
animal
elephant
zebras
Elephantidae
animals
methodology
sampling
boxes (containers)
aircraft
detection
animal population
method
population size
livestock
cameras
neural networks
human resources
wildlife
air

Keywords

  • computer vision
  • convolutional neural network
  • deep machine learning
  • drones
  • game census
  • image recognition
  • savanna
  • wildlife survey

Cite this

@article{b0b81bd8ac4d4dbd86dbf8a04294c7cf,
title = "Improving the precision and accuracy of animal population estimates with aerial image object detection",
abstract = "Animal population sizes are often estimated using aerial sample counts by human observers, both for wildlife and livestock. The associated methods of counting remained more or less the same since the 1970s, but suffer from low precision and low accuracy of population estimates. Aerial counts using cost-efficient Unmanned Aerial Vehicles or microlight aircrafts with cameras and an automated animal detection algorithm can potentially improve this precision and accuracy. Therefore, we evaluated the performance of the multi-class convolutional neural network RetinaNet in detecting elephants, giraffes and zebras in aerial images from two Kenyan animal counts. The algorithm detected 95{\%} of the number of elephants, 91{\%} of giraffes and 90{\%} of zebras that were found by four layers of human annotation, of which it correctly detected an extra 2.8{\%} of elephants, 3.8{\%} giraffes and 4.0{\%} zebras that were missed by all humans, while detecting only 1.6 to 5.0 false positives per true positive. Furthermore, the animal detections by the algorithm were less sensitive to the sighting distance than humans were. With such a high recall and precision, we posit it is feasible to replace manual aerial animal count methods (from images and/or directly) by only the manual identification of image bounding boxes selected by the algorithm and then use a correction factor equal to the inverse of the undercounting bias in the calculation of the population estimates. This correction factor causes the standard error of the population estimate to increase slightly compared to a manual method, but this increase can be compensated for when the sampling effort would increase by 23{\%}. However, an increase in sampling effort of 160{\%} to 1,050{\%} can be attained with the same expenses for equipment and personnel using our proposed semi-automatic method compared to a manual method. Therefore, we conclude that our proposed aerial count method will improve the accuracy of population estimates and will decrease the standard error of population estimates by 31{\%} to 67{\%}. Most importantly, this animal detection algorithm has the potential to outperform humans in detecting animals from the air when supplied with images taken at a fixed rate.",
keywords = "computer vision, convolutional neural network, deep machine learning, drones, game census, image recognition, savanna, wildlife survey",
author = "Eikelboom, {Jasper A.J.} and Johan Wind and {van de Ven}, Eline and Kenana, {Lekishon M.} and Bradley Schroder and {de Knegt}, {Henrik J.} and {van Langevelde}, Frank and Prins, {Herbert H.T.}",
year = "2019",
month = "11",
doi = "10.1111/2041-210X.13277",
language = "English",
volume = "10",
pages = "1875--1887",
journal = "Methods in Ecology and Evolution",
issn = "2041-210X",
publisher = "Wiley",
number = "11",

}

Improving the precision and accuracy of animal population estimates with aerial image object detection. / Eikelboom, Jasper A.J.; Wind, Johan; van de Ven, Eline; Kenana, Lekishon M.; Schroder, Bradley; de Knegt, Henrik J.; van Langevelde, Frank; Prins, Herbert H.T.

In: Methods in Ecology and Evolution, Vol. 10, No. 11, 11.2019, p. 1875-1887.

Research output: Contribution to journalArticleAcademicpeer-review

TY - JOUR

T1 - Improving the precision and accuracy of animal population estimates with aerial image object detection

AU - Eikelboom, Jasper A.J.

AU - Wind, Johan

AU - van de Ven, Eline

AU - Kenana, Lekishon M.

AU - Schroder, Bradley

AU - de Knegt, Henrik J.

AU - van Langevelde, Frank

AU - Prins, Herbert H.T.

PY - 2019/11

Y1 - 2019/11

N2 - Animal population sizes are often estimated using aerial sample counts by human observers, both for wildlife and livestock. The associated methods of counting remained more or less the same since the 1970s, but suffer from low precision and low accuracy of population estimates. Aerial counts using cost-efficient Unmanned Aerial Vehicles or microlight aircrafts with cameras and an automated animal detection algorithm can potentially improve this precision and accuracy. Therefore, we evaluated the performance of the multi-class convolutional neural network RetinaNet in detecting elephants, giraffes and zebras in aerial images from two Kenyan animal counts. The algorithm detected 95% of the number of elephants, 91% of giraffes and 90% of zebras that were found by four layers of human annotation, of which it correctly detected an extra 2.8% of elephants, 3.8% giraffes and 4.0% zebras that were missed by all humans, while detecting only 1.6 to 5.0 false positives per true positive. Furthermore, the animal detections by the algorithm were less sensitive to the sighting distance than humans were. With such a high recall and precision, we posit it is feasible to replace manual aerial animal count methods (from images and/or directly) by only the manual identification of image bounding boxes selected by the algorithm and then use a correction factor equal to the inverse of the undercounting bias in the calculation of the population estimates. This correction factor causes the standard error of the population estimate to increase slightly compared to a manual method, but this increase can be compensated for when the sampling effort would increase by 23%. However, an increase in sampling effort of 160% to 1,050% can be attained with the same expenses for equipment and personnel using our proposed semi-automatic method compared to a manual method. Therefore, we conclude that our proposed aerial count method will improve the accuracy of population estimates and will decrease the standard error of population estimates by 31% to 67%. Most importantly, this animal detection algorithm has the potential to outperform humans in detecting animals from the air when supplied with images taken at a fixed rate.

AB - Animal population sizes are often estimated using aerial sample counts by human observers, both for wildlife and livestock. The associated methods of counting remained more or less the same since the 1970s, but suffer from low precision and low accuracy of population estimates. Aerial counts using cost-efficient Unmanned Aerial Vehicles or microlight aircrafts with cameras and an automated animal detection algorithm can potentially improve this precision and accuracy. Therefore, we evaluated the performance of the multi-class convolutional neural network RetinaNet in detecting elephants, giraffes and zebras in aerial images from two Kenyan animal counts. The algorithm detected 95% of the number of elephants, 91% of giraffes and 90% of zebras that were found by four layers of human annotation, of which it correctly detected an extra 2.8% of elephants, 3.8% giraffes and 4.0% zebras that were missed by all humans, while detecting only 1.6 to 5.0 false positives per true positive. Furthermore, the animal detections by the algorithm were less sensitive to the sighting distance than humans were. With such a high recall and precision, we posit it is feasible to replace manual aerial animal count methods (from images and/or directly) by only the manual identification of image bounding boxes selected by the algorithm and then use a correction factor equal to the inverse of the undercounting bias in the calculation of the population estimates. This correction factor causes the standard error of the population estimate to increase slightly compared to a manual method, but this increase can be compensated for when the sampling effort would increase by 23%. However, an increase in sampling effort of 160% to 1,050% can be attained with the same expenses for equipment and personnel using our proposed semi-automatic method compared to a manual method. Therefore, we conclude that our proposed aerial count method will improve the accuracy of population estimates and will decrease the standard error of population estimates by 31% to 67%. Most importantly, this animal detection algorithm has the potential to outperform humans in detecting animals from the air when supplied with images taken at a fixed rate.

KW - computer vision

KW - convolutional neural network

KW - deep machine learning

KW - drones

KW - game census

KW - image recognition

KW - savanna

KW - wildlife survey

U2 - 10.1111/2041-210X.13277

DO - 10.1111/2041-210X.13277

M3 - Article

VL - 10

SP - 1875

EP - 1887

JO - Methods in Ecology and Evolution

JF - Methods in Ecology and Evolution

SN - 2041-210X

IS - 11

ER -