dc.description.abstract |
Since 1994 Lynx are under strict species protection and is
part of the Polish Red Book of Animals (Polska czerwona ksiega zwierzat),
therefore plenty of measures were taken to preserve and care for this pro-
tected species. A monitoring system, including several camera traps, of
wild animals in their natural habitat was expanded to include Lynx. This
system generates data of wildlife continuously and in large volume which
relies on citizen scientists to manually process images and videos cap-
tured from camera traps. The process can be extremely time consuming
and laborious. Leveraging the recent advances in deep learning techniques
and computer vision, we overcome this obstacle with the idea of devel-
oping an automated framework to automatically identify and recognize
Lynx in the wild. We kick o this project by using a sample dataset of
Lynx images from the National Park to nd the most accurate image
recognition model. First by assessing three di erent state-of-the-art deep
learning algorithms using Transfer Learning methods then by training
our own benchmark Convolutional Neural Network models using three
complementary architectures. The results later were compared and dis-
cussed taking into account the accuracy and speed of each method. In
turn, a proposal concerning the model with the best performance for fur-
ther testing and implementation was introduced. And nally we re
ect
on our work by structuring an improvement plan and planning future
stages in this project. |
pl_PL |