Wildlife Insights – using AI to identify and monitor wildlife populations

Wildlife Insights – using AI to identify and monitor wildlife populations

In this era of environmental challenges, a new project using artificial intelligence (AI) is helping to monitor and manage wildlife populations. Wildlife Insights claims to be “the largest and most diverse collection of camera trap images that is open to the public”.

The project utilises a number of tools, including artificial intelligence models, to identify varied species, generate automated statistics and features a cloud-based platform to facilitate the sharing of data among stakeholders.

Researchers are able to run AI models over their data on the Google Cloud-based platform. Through the use of the framework TensorFlow, Google has trained AI models to perform both blank image filtering and species classification. (TensorFlow is an end-to-end open-source platform used for machine learning). Wildlife Insights states that these AI models are able to detect nearly 79% of blank images with an error rate of under 2%.

A screenshot of part of Wildlife Insights' home page
A screenshot of part of Wildlife Insights’ home page

Open to the public, the portal currently contains around 4.5 million photos. Anyone is welcome to access the photos and assist by pinpointing locations. Collaborators are further invited to assist with growth of the database by uploading their own camera trap imagery.

As phenomena like grass blowing in the wind can trigger a camera trap, datasets can become massively swelled through imagery such as this. Leveraging Google AI Platform Predictions, automating the identification of these blank images, i.e. where no animals are present, the process of identifying data in the camera trap imagery is dramatically sped up, allowing researchers to focus on just those images they require or prefer – those containing identified species.

Here’s a video of the platform at work identifying uploaded images. Notice the scored options provided when the score is lower than hoped for and the AI is thus not able to give a near-certain identification.

The Wildlife Insights site answers the question of how the AI models decide on a particular class by describing the use of convolutional neural networks. Quoting verbatim, these networks are listed as:

a widely successful AI paradigm for computer vision models. At a high level, the model takes an image as a 2D input (array of pixels in single channel or RGB channels) and runs mathematical operations in a set of steps. Each step is referred to as a layer. There are some peculiar layer types used in CNNs for images, like convolution and pooling. Multiple such layers are collated together to form a deep convolutional neural network.

For further information on, and understanding of, convolution and pooling, or to learn more about some of the technology behind the platform, see the following articles on Towards Data Science and Hacker Noon.

Data can be shared with Wildlife Insights through their API and both eMammal and Wild.ID users will soon be able to sync with the platform.

The project has some serious backers too, with founding and core members including Conservation International, WWF, Google, Wildlife Conservation Society, Map of Life, Smithsonian’s National Zoo & Conservation Biology Institute, North Carolina Museum of Natural Sciences and the Zoological Society of London

This project has the hallmarks of being a great resource for the public, conservationists, academics and researchers too. If the number of images added continues to grow, the datasets in turn will too, and undoubtedly, so too the AI and ML capabilities. The current data can be viewed online directly at https://app.wildlifeinsights.org/explore.

Head over to Wildlife Insights’ website to check out the project here


References & Sources:

Funes, Y., 2019. Google’s New AI Project Could Be A Conservation Game Changer. [online] Gizmodo. Available at: https://earther.gizmodo.com/google-develops-ai-to-mine-camera-trap-photos-in-the-na-1840484176

Google Cloud. n.d. Prediction Overview | AI Platform Prediction | Google Cloud. [online] Available at: https://cloud.google.com/ai-platform/prediction/docs/overview

Reppel, E., 2017. Visualizing Parts Of Convolutional Neural Networks Using Keras And Cats | Hacker Noon. [online] Hackernoon.com. Available at: https://hackernoon.com/visualizing-parts-of-convolutional-neural-networks-using-keras-and-cats-5cc01b214e59

Saha, S., 2018. A Comprehensive Guide To Convolutional Neural Networks — The ELI5 Way. [online] Towards Data Science. Available at: https://towardsdatascience.com/a-comprehensive-guide-to-convolutional-neural-networks-the-eli5-way-3bd2b1164a53

TensorFlow. n.d. Tensorflow. [online] Available at: https://www.tensorflow.org/

Wildlifeinsights.org. n.d. Home | Wildlife Insights. [online] Available at: https://www.wildlifeinsights.org/


Keen to be notified of new posts? Sign up for Coding Climate’s newsletter:

Leave a Reply