Full text not available from this repository.

Abstract

In this paper, we present R-CNN, Fast R-CNN and Faster R-CNN methods to automatically detect and recognise the predators in underwater videos. We compare the results of these methods on real data and discuss their strengths and weaknesses. We build a dataset using footage captured from representative environment of the wild and devise a data model with three classes (seal, dolphin, background). Following this, we train R-CNN, Fast R-CNN and Faster R-CNN, then evaluate them on a test dataset compose of challenging objects that had not been seen during training. We perform evaluation on GPU, acquiring information about the AP and IOU for each model and network based on various proposal numbers as well as runtime speeds. Based on the results, we found that the best model of predator detection using visual deep learning models is Faster R-CNN with 2000 proposals.

Item Type:	Conference Publication
Authors/Creators:	Park, M and Yang, W and Cao, Z and Kang, B and Connor, D and Lea, M-A
Keywords:	R-CNN, fast R-CNN, faster R-CNN, marine vertebrate, seal, dolphin, detection, recognition, deep learning
Journal or Publication Title:	Lecture Note in Computer Science: Proceedings of the 16th Pacific Rim Knowledge Acquisition Workshop: Knowledge Management and Acquisition for Intelligent Systems (PKAW 2019)
Publisher:	Springer
Copyright Information:	Copyright 2019 Springer
Related URLs:	Google Scholar: Cao, Z Google Scholar: Kang, B UTAS Profile: Kang, B Google Scholar: Lea, M-A UTAS Profile: Lea, M-A
Item Statistics:	View statistics for this item

Actions (login required)

Item Control Page