Over the last century, the population of African elephants (Loxodonta africana) has declined sharply due to poaching.
Currently the most commonA method of surveying elephant populations in the savannah is aerial counting from manned aircraft. Observers involved in aerial photography can face problems due to poor visibility, and aerial photography itself can be costly and logistically difficult.
A team from Oxford University (WildCRU:Faculty of Zoology and Machine Learning Research Group: Faculty of Engineering), in collaboration with Dr. Olga Isupova, University of Bath, and Dr. Tiejun Wang, University of Twente, set out to address these challenges.
Remote sensing helped withsatellite imagery and automation of elephant detection using deep learning. The new method solves various existing problems of population control. Satellites can collect images over 5000 km² in a single pass in minutes, eliminating the risk of double counting.
Satellite monitoring is an unobtrusive method,does not require a presence on the ground, which eliminates the risk of interference in the life of populations or a threat to human safety during data collection. Previously inaccessible areas become accessible, and border areas, often critical for conservation planning, can be surveyed without time-consuming ground permitting requirements.
One of the problems of using satellitemonitoring - processing of a huge number of created images. However, automating discovery cuts the processing time from months to seconds. In addition, the machines are less error prone.
To develop this new method, the team created a custom training dataset of over 1,000 elephants in South Africa, which was fed into a convolutional neural network (CNN).
Researchers are confident that satellite remote sensing and deep learning technologies will help preserve these majestic mammals.
Read also
AI solved the Schrödinger equation
Abortion and science: what will happen to the children who will give birth
"Study failed": Sputnik V testers will no longer receive placebo