Deep object classification in low resolution LWIR imagery via transfer learning

R. Abbott; J. M.  Del Rincon; B. Connor; N. Robertson

Deep object classification in low resolution LWIR imagery via transfer learning

R. Abbott, J. M. Del Rincon, B. Connor, N. Robertson

School of Electronics, Electrical Engineering and Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

1162 Downloads (Pure)

Abstract

In this paper we present an object detection system based on YOLOv2 and transfer learning which is compared to Keras faster RCNN. We evaluate our systems using two infrared datasets: a small dataset of low resolution thermal images and a large dataset of high resolution thermal images both containing the object classes; people and land vehicles. Both detectors are trained on the large dataset of high resolution thermal images and tested using the low resolution images. Fine tuning on the small dataset is implemented to increase the accuracy of the detectors. This research will be of great interest to the defence community as they could save a lot of time and money collecting and annotating data.

Original language	English
Title of host publication	Proceedings of the 5th IMA Conference on Mathematics in Defence
Number of pages	9
Publication status	Published - 23 Nov 2017

Access to Document

Deep object classification in low resolution LWIR imagery via transfer learning
© 2017 The Institute of Mathematics and its Applications. This work is made available online in accordance with the publisher’s policies. Please refer to any applicable terms of use of the publisher.
Accepted author manuscript, 4.31 MBLicence: Unspecified

Multi-modal object detection in non-corresponding imagery using unsupervised techniques
Abbott, R. (Author), Martinez del Rincon, J. (Supervisor) & Robertson, N. (Supervisor), Jul 2021
Student thesis: Doctoral Thesis › Doctor of Philosophy

File

Cite this

@inproceedings{998c93e040c941dcaf646b2663348225,

title = "Deep object classification in low resolution LWIR imagery via transfer learning",

abstract = "In this paper we present an object detection system based on YOLOv2 and transfer learning which is compared to Keras faster RCNN. We evaluate our systems using two infrared datasets: a small dataset of low resolution thermal images and a large dataset of high resolution thermal images both containing the object classes; people and land vehicles. Both detectors are trained on the large dataset of high resolution thermal images and tested using the low resolution images. Fine tuning on the small dataset is implemented to increase the accuracy of the detectors. This research will be of great interest to the defence community as they could save a lot of time and money collecting and annotating data.",

author = "R. Abbott and {Del Rincon}, {J. M.} and B. Connor and N. Robertson",

year = "2017",

month = nov,

day = "23",

language = "English",

booktitle = "Proceedings of the 5th IMA Conference on Mathematics in Defence",

}

TY - GEN

T1 - Deep object classification in low resolution LWIR imagery via transfer learning

AU - Abbott, R.

AU - Del Rincon, J. M.

AU - Connor, B.

AU - Robertson, N.

PY - 2017/11/23

Y1 - 2017/11/23

N2 - In this paper we present an object detection system based on YOLOv2 and transfer learning which is compared to Keras faster RCNN. We evaluate our systems using two infrared datasets: a small dataset of low resolution thermal images and a large dataset of high resolution thermal images both containing the object classes; people and land vehicles. Both detectors are trained on the large dataset of high resolution thermal images and tested using the low resolution images. Fine tuning on the small dataset is implemented to increase the accuracy of the detectors. This research will be of great interest to the defence community as they could save a lot of time and money collecting and annotating data.

AB - In this paper we present an object detection system based on YOLOv2 and transfer learning which is compared to Keras faster RCNN. We evaluate our systems using two infrared datasets: a small dataset of low resolution thermal images and a large dataset of high resolution thermal images both containing the object classes; people and land vehicles. Both detectors are trained on the large dataset of high resolution thermal images and tested using the low resolution images. Fine tuning on the small dataset is implemented to increase the accuracy of the detectors. This research will be of great interest to the defence community as they could save a lot of time and money collecting and annotating data.

M3 - Conference contribution

BT - Proceedings of the 5th IMA Conference on Mathematics in Defence

ER -

Deep object classification in low resolution LWIR imagery via transfer learning

Abstract

Access to Document

Fingerprint

Student theses

Multi-modal object detection in non-corresponding imagery using unsupervised techniques

Cite this