Saturday, January 23

Elephants can now be surveyed from space


Elephant counting with satellite images

Elephant counting with satellite images
Oxford University

Using the highest resolution currently available in satellite imagery – Worldview 3 from Maxar Technologies – and deep learning, Oxford researchers have detected elephants from space with an accuracy comparable to that of human detection capabilities.

The population of African elephants (Loxodonta africana) has plummeted over the last century due to the poaching, the killing in retaliation for theft of crops and habitat fragmentation. Keeping them requires knowing where they are and how many there are – accurate tracking is vital.

Existing methods are prone to errors. The inaccurate counts lead to misallocation of scarce conservation resources and a poor understanding of population trends.

Currently, the most common survey technique for elephant populations in savanna environments is the aerial counting from manned aircraft. Observers in aerial reconnaissance can become exhausted, hampered by poor visibility and succumb to bias, and aerial reconnaissance can be costly and logistically challenging. A team from the University of Oxford (WildCRU: Department of Zoology and Machine Learning Research Group: Department of Engineering) set out to solve these problems.

Remote elephant detection using satellite imagery and automatic detection through deep learning provides a new method for studying elephants and also solves several existing challenges. Satellites can collect more than 5,000 square kilometers of images in a single pass captured in minutes, eliminating the risk of double counting. It is also possible to repeat the surveys at short intervals.

Satellite monitoring is a discreet technique that requires no ground presence, eliminating the risk of disturbing species or concerns for human safety during data collection. Areas that were previously inaccessible become accessible, and transboundary areas, often crucial for conservation planning, can be inspect without requiring time-consuming ground permits, reports the University of Oxford in a statement.

One of the challenges of using satellite monitoring is processing the huge amount of images generated. However, automating detection means that a process that would formally have taken months can be completed in a matter of hours. Further, machines are less prone to errors, the false negatives and false positives in deep learning algorithms are consistent and can be rectified by systematically improving the models; The same cannot be said for humans.

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) and the results were compared with human performance. It turns out that elephants can be detected in satellite images with as high a precision as human detection ability. The results (known as the F2 score) of the CNN models were 0.78 in heterogeneous areas and 0.73 in homogeneous areas, compared to an average human detection capacity F2 score of 0.77 in heterogeneous areas and 0.80 in homogeneous areas. The model could even detect elephants in locations far from the training data site, which shows the generalization of the model. Having trained the machine only in adults, he was able to identify the young.

The researchers believe this demonstrates the power of technology for conservation: satellite remote sensing and deep learning technologies promise the conservation of these majestic mammals. Conservation technologies open a new world of possibilities, which will be embraced with the urgency required by the sixth mass extinction and the difficult global situation of biodiversity.

El estudio se publica en Remote Sensing in Ecology and Conservation.

style="display:block" data-ad-client="ca-pub-3066188993566428" data-ad-slot="4073357244" data-ad-format="auto" data-full-width-responsive="true">
www.informacion.es

Leave a Reply

Your email address will not be published. Required fields are marked *

LinkedIn
Share