The model uses an aerial image to accelerate the survey. Credit: NIOZ
Scientists at EPFL, the Royal Netherlands Institute for Sea Research and Wageningen University & Research have developed a new model for counting the number of seals in aerial photos that is considerably faster than doing it by hand. Valuable time and resources can be saved with this new method.
Ecologists have been taking aerial photos of seal populations for decades. It takes hours of labor to manually identify the animals in each image.
A team of researchers, including a PhD student and an associate professor, are working together. They used a deep-learning model to count the number of seals in the photos. Their method could run through 100 images in less than an hour.
No labeling is needed.
The most commonly used deep-learning models are trained to detect individual objects after which they are counted. "This type of model requires extensive annotations of individual objects during training." The method used by the research team eliminates the need to label individual seals, which is a huge help since only the total number of animals in the picture is needed. Their method can be used to count any items or individual animals, and thus potentially help to process not only the new photos, but also those that could not be analyzed for lack of time. This represents decades of photos that could provide insight into how population size has changed over time.
From the smallest to the largest.
Depending on the altitude and angle at which the photo was taken, the way seals appear in aerial photos can be vastly different. The research team evaluated the robustness of the variation. To demonstrate the potential of their deep-learning model, the scientists tested their approach on a fundamentally different dataset of images of tiny growth rings in fish bones. These hearing stones are located behind a fish's brain. The daily growth rings visible in the images are used to estimate the age of the fish by the scientists. These growth rings are very difficult to write about. The research team found that their model had the same margin of error as manual methods, but could work through 100 images in under a minute, whereas it would take three hours for an expert.
The next step is to use satellite images of the remote areas of the north where several seal populations live that are on the Red List of Threatened Species. Tuia says that they plan to use this approach to study the plight of some of the world's most vulnerable species. Knowing where the animals concentrate is important to protect them.
The value of ecological surveys is given by counting using deep learning regression. The DOI is 10.1038/s41598-021-02387-9.
Scientific reports are in the journal.
Artificial intelligence helps speed up ecological surveys.
The document is copyrighted. Any fair dealing for the purpose of private study or research cannot be reproduced without written permission. The content is not intended to be used for anything other than information purposes.