Calling all Brown Bear Biologists/Managers/Enthusiasts - Help us to Develop a New Method to Recognise and Monitor Brown Bears Worldwide! BearlD: Developing Face Recognition Technology for Brown Bears using Al (2017-)

摘要

In 2017 the Brown Bear Research Network is launching a new research project in which we are developing a software tool which has the ability to recognise individual brown bears from images and video footage of their faces. Using cutting-edge deep-learning methods of computer learning, we are designing a system which can recognise individual bears with an accuracy of over 80%. Deep-learning methods require large datasets of images, as the computer system trains itself on pattern recognition. Therefore, thousands of training images are required to give the high accuracy of recognition which we require. A pilot study funded by Knight Inlet Lodge (a bear-viewing lodge in British Columbia) in 2015/16 revealed the ability to recognise individual bears by their faces, with an accuracy of 69%. Switching to a deep-learning approach with an increased sample size should give us the accuracy we require to develop an open-source useable end-product. In 2017 we are using camera traps to collect as many images/videos as possible of recognisable, individual grizzly bears to add to our database.