Digital Conservation of the Cranes

Every fall, Cheorwon welcomes the cranes which fly several hundred kilometers from Siberia to Korea. After the harvest, large quantities of the grains are left in the rice paddies for the cranes to feed on, while the wetlands inside of the DMZ provide a safe resting place for them to sleep.

There are currently 14 species of cranes worldwide. Seven species of them are observed in the Korean Peninsula. Approximately 1,000 red-crowned cranes and 5,500 white-naped cranes (representing over 50 percent of the global population of their respective species) spend their winters in Cheorwon.

The cranes in Cheorwon are designated as internationally endangered species. Developmental projects near the CCL and the changes in farming methods poses additional threats to their habitats. Local residents in Cheorwon and researchers have been monitoring the crane population and their habitats. 

For effective crane survey, it is crucial to be equipped with: a sturdy vehicle that can travel over rough farm roads, a satellite map, a GPS navigation app, and the trained eyes to spot the cranes from far away!

The size and locations of the cranes are marked down on the print-outs of satellite maps. The red-crowned cranes are represented as triangles, and the white-naped cranes as circles. Cranes usually travel in family groups of three or four.

After the counting, the number of cranes per section, as recorded on the satellite map, are transcribed into the field notebooks, and then aggregated to calculate the total crane population. In January 2021, there was a total of 5,330 white-naped cranes and 1,126 red-crowned cranes.

Over the past two winters, tens of thousands of photos and videos were captured. Not being disturbed by any human presence or activities, the cranes look as if they are not being observed at all. Shall we take a look at the examples? 

In order to develop AI, the computer must be fed with the data that is correctly labeled. For the crowd counting of human population, the images are typically labelled by tagging a dot on the head of a person. However, this strategy was not easily applied to cranes; A lot of cranes feed with their head tucked under their wing, making it impossible to dot each head.

Then, how to label the cranes whose heads aren’t readily visible? The research team decided to draw a line that connects a dot on each crane’s head with a dot on their body. To reflect the crane’s unique characteristics and habits, the labeling method was changed from dots to lines.

Once the data is labeled, it is used for AI training. A computer is shown an image of the cranes, and provides an estimate of the crane population. Then the computer’s estimate is compared with the actual, correct figure. If the computer gives an incorrect answer, then it is sent back to square one! Repeating this process dozens of times, the computer finally begins to give much accurate answers. 

Let’s see how it works. This photo features dozens of white-naped cranes. The cranes with a brownish tinge on their head or neck are younger ones less than one year old. Those birds with a white head and neck are adults.

The AI is still much less capable than a human surveyor. Although it produces relatively accurate counts for simple photos, its accuracy level drops significantly for photos of multiple crane species. There is still a long way to go. For the continued development of AI ecologist, we will develop an interspecies partnership between humans and AI, rather than replacement of one by another, through which the human and AI surveyors can make the best of their capacities.