Scientists of the University of Oxford and Bath have successfully used Deep Learning successfully with colleagues from Twente to capture elephants in their natural environment on satellite images. The software could pay the animals with an accuracy equal or coarse than the human experts. For nature conservant, this is an important step in monitoring populations of hazardous species.
So far, the population of African elephants is usually determined by the evaluation of aerial photography with aircraft or drones. However, this method is made of several groups very mistaken, write Isla Duorge and her colleagues in the journal Remote Sensing in Ecology and Conversation. For on the one hand, from time-consuming and costly not the entire area to be observed, but only a part to increase the paid animals. On the other hand, there are always duplications – human experts also made mistakes.
Well hidden dickhauter
As a matter of principle, with satellite images, a lot of large areas can be monitored. The transmission of established methods for image processing on satellite images, however, has been promoted in a targeted for several years – for example for those inventive natural disasters. The observation of wild animals with the help of satellites has so far scanned optically simpler cases, such as whales in the sea, albatrosses or penguin colonies in the Antarctic.
However, the elephants analyzed in this study move through Walder and Grassland, and, among other things, have the habit of sousting in the mud, "otherwise continuing color and shape", write the researchers.
An prefeaked convolutional network that trained the researchers with high-minded images of the satellites Worldview 3 and 4, but with comparatively few examples was able to capture the African elephants. In the pictures, details are to be distinguished up to 31 centimeters roughly.
1600 square kilometers
The population of the African elephants was heavily inflicted in the last century mainly due to poaching and the dull of the habitat. With about 415.000 African savanna elephants who still live in the wild, the species is classified as danger.
"Accurate monitoring is unasked if we want to save the species", says Olga Isupova from the University of Bath, which has programmed the neuronal network. "We have to know where the animals are and how many are."
This process, which usually lasts weeks, can now be completed in a few hours. However, a disadvantage is the cost of the pictures: An already existing image costs $ 17.50 per quadrakilometer – a newly to be recorded $ 27.50. The National Park supervised in this study has a flat of about 1600 square kilometers.