One of the key questions driving our research is how wildlife and computer vision technology (such as deep learning for image processing) intersect. Specifically, when deep learning and computer vision engage with wild places, ecosystems, and species, what classification errors, boundary issues, technical limitations, or anthropocentric biases might surface?

To explore this, we spoke with ecological researcher Woo Donggul from the National Institute of Ecology, as well as activists Park Eunjung and Park Sung Jun from Green Korea United. Their insights on how technology is being used to protect and conserve wildlife helped shape the foundation of our research. Below are highlights from our preliminary consultations and interviews. (We extend our gratitude to these experts for sharing their knowledge and perspectives.)

# Wildlife Survey Technology – Camera Trap Use Cases

One of the most widely adopted technologies in wildlife research is the unmanned camera, commonly referred to in ecological studies as “camera trap surveys.” These cameras use motion sensors to capture video or photos of passing wildlife, allowing researchers to monitor species presence, activity, and population sizes.

Institutions like the National Ecological Center utilize unmanned cameras through a grid survey system. Survey areas are divided into grids of a set size, and cameras are placed in each grid to estimate wildlife density and numbers. The size of each grid is determined by the survey’s goals and the typical range of animal activity. This statistical sampling method helps ensure objectivity in data collection.

Green United Korea, an environmental organization, also employs unmanned cameras but supplements this with hands-on fieldwork. Instead of relying solely on statistical sampling, their approach involves tracking animal signs—like scat, footprints, or feeding patterns—and observing environmental disruptions such as habitat loss from urban development, wildfires, or heavy snowfall linked to the climate crisis. The combination of video data and direct observation helps Green United Korea provide empirical evidence of species presence and environmental impact.

Whereas public institutions focus on quantitative analysis of wildlife populations, Green United Korea emphasizes the broader relationship between human activities and wildlife habitats. Though their methods differ, both approaches are essential for collecting data and evidence crucial to wildlife conservation and protection.

# Wildlife Survey Technology – From Data to Insights

Grid surveys conducted with unmanned cameras provide a structured and quantitative means of data collection, enabling researchers to calculate wildlife density and activity levels. However, the data captured in the wild often reflects nature’s rhythms and animal behaviors that diverge from human perception and experience.

Over time, cameras gather vast amounts of footage, but because wildlife movements follow patterns distinct from human understanding, some recordings may appear irrelevant—essentially “noise” in the dataset. Both humans and machines might filter out this noise, excluding it from statistical samples.

Green United Korea’s conservation work, in contrast, involves actively collecting overlooked but meaningful evidence—like antler rubs on tree or animal droppings—that drone cameras or automated systems might miss. This labor-intensive process fills the gaps left by purely quantitative approaches, highlighting the importance of data that might otherwise go unrecognized.

Ultimately, both methodologies aim to protect wildlife, but they reflect different interpretations and applications of the data collected.

# Wildlife Survey Technology – Datasets and Machine Learning

For years, ecological researchers manually sifted through video footage from unmanned cameras to extract meaningful insights and compile species data. This process is highly labor-intensive, requiring significant time and effort.

Recently, computer vision and machine learning have emerged as effective tools to streamline this work. Deep learning models can analyze large datasets, identifying and classifying wildlife with greater speed and accuracy, freeing researchers to focus on more critical aspects of conservation.

Image recognition technology identifies patterns and features in wildlife footage, automating the classification of species and individuals. However, for this to work effectively, large volumes of high-quality data are essential. One persistent challenge is data imbalance—where certain species are underrepresented in datasets. For example, endangered animals like the mountain goat appear infrequently compared to more common species, resulting in long-tailed distributions within datasets.

Machine learning models trained on imbalanced data tend to prioritize species with higher representation, reducing overall accuracy for rare species. To address this, researchers adjust training datasets by applying weight to underrepresented species or artificially increasing the amount of rare data through augmentation techniques. Although computational methods have their limits, human creativity and interpretation play a vital role in bridging the gap between technological constraints and ecological complexities.

# Bridging Technology and Ecology

The integration of AI and other advanced technologies into wildlife research is accelerating, though this field remains relatively small compared to human-centered AI industries. These technologies enhance the efficiency and accuracy of ecological surveys, allowing researchers to process larger datasets and refine their understanding of wildlife.

However, focusing solely on efficiency can sometimes lead to the exclusion of important nuances and elements that resist easy classification. By recognizing both the potential and limitations of these technologies, we aim to explore how humans engage with ecosystems, wildlife, and conservation through ongoing research and collaboration.