AI-powered Photo Monitoring: Innovative Technology for Conservation Implemented in Bosque Pehuén

MegaDetector is an artificial intelligence tool that classifies images captured by camera traps that can help to design actions promoting the conservation of wildlife species. For about a year, this software has been used in Bosque Pehuén, a privately managed conservation area in the Andean region of La Araucanía, with the aim of improving data processing efficiency and expediting conservation decision-making.

Photo monitoring through camera traps allows the study of wildlife behavior in a specific area. However, there are challenges in processing and analyzing data due to the high volume of images generated, manually processed by conservation professionals and researchers. There is a global trend towards processing this information using innovative software trained with artificial intelligence to reduce analysis times and enhance management efforts and decision-making in nature conservation.

In Chile, since 2022, the National Forestry Corporation (CONAF) started implementing artificial intelligence in 35 protected wild areas to enhance the data processing system. This innovation, first introduced in the country, automates the current manual classification process performed by CONAF professionals, saving 99% of the time spent by state protected areas officials reviewing camera trap material.

Cámara trampa- Bosque Pehuén

In the case of Bosque Pehuén, we have joined this initiative by using the MegaDetector software – currently under evaluation –, an image recognition software, free and openly available, created by a team from the California Institute of Technology and Microsoft AI for Nature. This model, trained with millions of images from around the world, is designed to detect humans, animals, and vehicles, facilitating the filtering of blank or “ghost images” captured when cameras react to movements caused by external factors such as wind or shadows, generating large volumes of images, sometimes exceeding 70% of the captured images.

Monitoring for Conservation

Through the strategic installation of 17 camera traps in different geographic points of Bosque Pehuén, the current state of fauna, the presence, distribution, and density of vertebrate species inhabiting this territory, including mammals such as pumas, güiñas, and culpeo foxes, can be assessed. This information enables the development of effective initiatives and measures for biodiversity conservation against potential threats.
The cameras are equipped with motion sensors that activate and capture images of individuals passing through their field of view. Once the information is collected, data processing is usually carried out manually: reviewing each image, classifying by species and geographic location, among other predefined data. This is where the use of artificial intelligence technologies marks a significant improvement in the time dedicated to this task, becoming tools that enhance and expedite the processing of information for conservation purposes.

How does AI work in species monitoring?

Although this image classification tool can have significant effects on reducing the time spent on data processing, it must be trained beforehand. Deep learning uses artificial neural networks to analyze information. These networks are based on or “mimic” biological neural networks, allowing algorithms to learn based on training data. In this case, the software must be fed with large amounts of correctly classified images to ensure its accuracy in detection.

Currently, our conservation team is in the evaluation stage of using this software, providing images corresponding to ecosystems in southern Chile, through photos taken in Bosque Pehuén, to analyze how effective this tool is at recognizing specific characteristics associated with the species inhabiting the protected area. According to Felipe Guarda, in charge of evaluation and studies at FMA, it is necessary to assess the performance of AI programs, as their performance depends on the databases they are trained on, in this case, mostly from the global north. “Good performance of Megadetector in those northern ecosystems may not transfer to our context, so it is important to analyze its performance before formally incorporating it into our image processing protocols,” he adds.

This tool allows accelerating the workflow, obtaining real-time information, and focusing protection efforts on the conservation objects of the area. It also allows focusing on the identification and control of threats, including, for example, human intrusion or the presence of invasive fauna species affecting local biodiversity, such as wild boars or American minks.
By 2024, the conservation team will continue working on the implementation of these technologies and the analysis of image content, conducting data crossovers to determine the abundance of certain key species, richness, spatial distribution, and temporal activity. Additionally, there are plans to modify the current monitoring network by redistributing the location of camera traps to achieve greater coverage of the territory and complement the monitoring with sound recorders (songmeters) for the study of other animals such as amphibians or birds, for transdisciplinary research purposes.

Soon, as part of a collaboration agreement signed in 2020 with CONAF La Araucanía Region, our foundation will seek to generate collaborative work to enhance data crossover recorded by camera traps between Pehuén Forest and other protected areas in the region, to study the fauna inhabiting these mountain biological corridors.