Performance comparison of deep learning techniques for recognizing birds in aerial images

Yang Liu, Peng Sun, Max R. Highsmith, Nickolas M. Wergeles, Joel Sartwell, Andy Raedeke, Mary Mitchell, Heath Hagy, Andrew D. Gilbert, Brian Lubinski, Yi Shang

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In computer vision, significant advances have been made in recent years on object recognition and detection with the rapid development of deep learning, especially deep convolutional neural networks (CNN). The majority of deep learning methods for object detection have been developed for large objects and their performances on small-object detection are not very good. This paper contributes to research in low-resolution small-object detection by evaluating the performances of leading deep learning methods for object detection using a common dataset, which is a new dataset for bird detection, called Little Birds in Aerial Imagery (LBAI), created from real-life aerial imagery data. LBAI contains birds with sizes ranging from 10px to 40px. In our experiments, some of the best deep learning architectures were implemented and applied to LBAI, which include object detection techniques such as YOLOv2, SSH, and Tiny Face, in addition to small instance segmentation techniques including U-Net and Mask R-CNN. Among the object detection methods, experimental results demonstrated that SSH performed the best for easy cases, whereas Tiny Face performed the best for hard cases, i.e. where a cluttered background makes detecting birds difficult. Among small instance segmentation methods, experimental results revealed U-Net achieved slightly better performance than Mask R-CNN.

Original languageEnglish (US)
Title of host publicationProceedings - 2018 IEEE 3rd International Conference on Data Science in Cyberspace, DSC 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages317-324
Number of pages8
ISBN (Electronic)9781538642108
DOIs
StatePublished - Jul 16 2018
Event3rd IEEE International Conference on Data Science in Cyberspace, DSC 2018 - Guangzhou, Guangdong, China
Duration: Jun 18 2018Jun 21 2018

Conference

Conference3rd IEEE International Conference on Data Science in Cyberspace, DSC 2018
CountryChina
CityGuangzhou, Guangdong
Period6/18/186/21/18

Fingerprint

Birds
Antennas
Neural networks
Masks
Object recognition
Deep learning
Object detection
Computer vision
Imagery

Keywords

  • INHS
  • Convolutional neural networks
  • Deep learning
  • Instance segmentation
  • Small-object detection
  • Aerial image dataset

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Information Systems
  • Information Systems and Management
  • Safety, Risk, Reliability and Quality

Cite this

Liu, Y., Sun, P., Highsmith, M. R., Wergeles, N. M., Sartwell, J., Raedeke, A., ... Shang, Y. (2018). Performance comparison of deep learning techniques for recognizing birds in aerial images. In Proceedings - 2018 IEEE 3rd International Conference on Data Science in Cyberspace, DSC 2018 (pp. 317-324). [8411873] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/DSC.2018.00052

Performance comparison of deep learning techniques for recognizing birds in aerial images. / Liu, Yang; Sun, Peng; Highsmith, Max R.; Wergeles, Nickolas M.; Sartwell, Joel; Raedeke, Andy; Mitchell, Mary; Hagy, Heath; Gilbert, Andrew D.; Lubinski, Brian; Shang, Yi.

Proceedings - 2018 IEEE 3rd International Conference on Data Science in Cyberspace, DSC 2018. Institute of Electrical and Electronics Engineers Inc., 2018. p. 317-324 8411873.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Liu, Y, Sun, P, Highsmith, MR, Wergeles, NM, Sartwell, J, Raedeke, A, Mitchell, M, Hagy, H, Gilbert, AD, Lubinski, B & Shang, Y 2018, Performance comparison of deep learning techniques for recognizing birds in aerial images. in Proceedings - 2018 IEEE 3rd International Conference on Data Science in Cyberspace, DSC 2018., 8411873, Institute of Electrical and Electronics Engineers Inc., pp. 317-324, 3rd IEEE International Conference on Data Science in Cyberspace, DSC 2018, Guangzhou, Guangdong, China, 6/18/18. https://doi.org/10.1109/DSC.2018.00052
Liu Y, Sun P, Highsmith MR, Wergeles NM, Sartwell J, Raedeke A et al. Performance comparison of deep learning techniques for recognizing birds in aerial images. In Proceedings - 2018 IEEE 3rd International Conference on Data Science in Cyberspace, DSC 2018. Institute of Electrical and Electronics Engineers Inc. 2018. p. 317-324. 8411873 https://doi.org/10.1109/DSC.2018.00052
Liu, Yang ; Sun, Peng ; Highsmith, Max R. ; Wergeles, Nickolas M. ; Sartwell, Joel ; Raedeke, Andy ; Mitchell, Mary ; Hagy, Heath ; Gilbert, Andrew D. ; Lubinski, Brian ; Shang, Yi. / Performance comparison of deep learning techniques for recognizing birds in aerial images. Proceedings - 2018 IEEE 3rd International Conference on Data Science in Cyberspace, DSC 2018. Institute of Electrical and Electronics Engineers Inc., 2018. pp. 317-324
@inproceedings{f5749837f9724b71b70a2373c52625c1,
title = "Performance comparison of deep learning techniques for recognizing birds in aerial images",
abstract = "In computer vision, significant advances have been made in recent years on object recognition and detection with the rapid development of deep learning, especially deep convolutional neural networks (CNN). The majority of deep learning methods for object detection have been developed for large objects and their performances on small-object detection are not very good. This paper contributes to research in low-resolution small-object detection by evaluating the performances of leading deep learning methods for object detection using a common dataset, which is a new dataset for bird detection, called Little Birds in Aerial Imagery (LBAI), created from real-life aerial imagery data. LBAI contains birds with sizes ranging from 10px to 40px. In our experiments, some of the best deep learning architectures were implemented and applied to LBAI, which include object detection techniques such as YOLOv2, SSH, and Tiny Face, in addition to small instance segmentation techniques including U-Net and Mask R-CNN. Among the object detection methods, experimental results demonstrated that SSH performed the best for easy cases, whereas Tiny Face performed the best for hard cases, i.e. where a cluttered background makes detecting birds difficult. Among small instance segmentation methods, experimental results revealed U-Net achieved slightly better performance than Mask R-CNN.",
keywords = "INHS, Convolutional neural networks, Deep learning, Instance segmentation, Small-object detection, Aerial image dataset",
author = "Yang Liu and Peng Sun and Highsmith, {Max R.} and Wergeles, {Nickolas M.} and Joel Sartwell and Andy Raedeke and Mary Mitchell and Heath Hagy and Gilbert, {Andrew D.} and Brian Lubinski and Yi Shang",
year = "2018",
month = "7",
day = "16",
doi = "10.1109/DSC.2018.00052",
language = "English (US)",
pages = "317--324",
booktitle = "Proceedings - 2018 IEEE 3rd International Conference on Data Science in Cyberspace, DSC 2018",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
address = "United States",

}

TY - GEN

T1 - Performance comparison of deep learning techniques for recognizing birds in aerial images

AU - Liu, Yang

AU - Sun, Peng

AU - Highsmith, Max R.

AU - Wergeles, Nickolas M.

AU - Sartwell, Joel

AU - Raedeke, Andy

AU - Mitchell, Mary

AU - Hagy, Heath

AU - Gilbert, Andrew D.

AU - Lubinski, Brian

AU - Shang, Yi

PY - 2018/7/16

Y1 - 2018/7/16

N2 - In computer vision, significant advances have been made in recent years on object recognition and detection with the rapid development of deep learning, especially deep convolutional neural networks (CNN). The majority of deep learning methods for object detection have been developed for large objects and their performances on small-object detection are not very good. This paper contributes to research in low-resolution small-object detection by evaluating the performances of leading deep learning methods for object detection using a common dataset, which is a new dataset for bird detection, called Little Birds in Aerial Imagery (LBAI), created from real-life aerial imagery data. LBAI contains birds with sizes ranging from 10px to 40px. In our experiments, some of the best deep learning architectures were implemented and applied to LBAI, which include object detection techniques such as YOLOv2, SSH, and Tiny Face, in addition to small instance segmentation techniques including U-Net and Mask R-CNN. Among the object detection methods, experimental results demonstrated that SSH performed the best for easy cases, whereas Tiny Face performed the best for hard cases, i.e. where a cluttered background makes detecting birds difficult. Among small instance segmentation methods, experimental results revealed U-Net achieved slightly better performance than Mask R-CNN.

AB - In computer vision, significant advances have been made in recent years on object recognition and detection with the rapid development of deep learning, especially deep convolutional neural networks (CNN). The majority of deep learning methods for object detection have been developed for large objects and their performances on small-object detection are not very good. This paper contributes to research in low-resolution small-object detection by evaluating the performances of leading deep learning methods for object detection using a common dataset, which is a new dataset for bird detection, called Little Birds in Aerial Imagery (LBAI), created from real-life aerial imagery data. LBAI contains birds with sizes ranging from 10px to 40px. In our experiments, some of the best deep learning architectures were implemented and applied to LBAI, which include object detection techniques such as YOLOv2, SSH, and Tiny Face, in addition to small instance segmentation techniques including U-Net and Mask R-CNN. Among the object detection methods, experimental results demonstrated that SSH performed the best for easy cases, whereas Tiny Face performed the best for hard cases, i.e. where a cluttered background makes detecting birds difficult. Among small instance segmentation methods, experimental results revealed U-Net achieved slightly better performance than Mask R-CNN.

KW - INHS

KW - Convolutional neural networks

KW - Deep learning

KW - Instance segmentation

KW - Small-object detection

KW - Aerial image dataset

UR - http://www.scopus.com/inward/record.url?scp=85051002258&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85051002258&partnerID=8YFLogxK

U2 - 10.1109/DSC.2018.00052

DO - 10.1109/DSC.2018.00052

M3 - Conference contribution

AN - SCOPUS:85051002258

SP - 317

EP - 324

BT - Proceedings - 2018 IEEE 3rd International Conference on Data Science in Cyberspace, DSC 2018

PB - Institute of Electrical and Electronics Engineers Inc.

ER -