This project began with the Con X Tech Competition hosted by Conservation X Labs, where 20 finalists were given a $3500 grant to address a conservation problem with a tech solution. SnapCat was pitched as a way to help researchers go through thousands of images with the help of computer vision, thereby reducing the time and money spent by these organizations.
The first case-study of SnapCat will be in partnership with the nonprofit Island Conservation. Island Conservation aims to protect island ecosystems by actively removing invasive species, thereby reducing the number of extinctions in the most biodiverse places of our planet. Island Conservation reports that 75% of all bird, amphibian, mammal, and reptile extinctions have occurred on islands. They have also underscored that 86% of all extinctions that were linked to invasives also occurred on islands.*
Several questions were posed to Conservation X Labs. How do we determine that all invasives have been removed from an island? How do we sort through large datasets of images to determine if a camera has seen an invasive? Is it possible to automatically detect for an invasive? Is it possible to have a real-time alert system tied to invasive detections?
SnapCat addresses these questions.
SnapCat isn't anything new or fancy. Our solution makes use of advancements in machine learning, low-power embedded systems, and wireless technologies. We aren't trying to make the tech better, we're just making it more accessible to the groups who need it.
SnapCat is based on the open-source TensorFlow software provided by Google. We make use of a pre-trained neural network that can be retrained to detect for a species (or thing!) of your choice. Our current initiative is to provide the nonprofit, Island Conservation, with a feral cat classifier that can be used in the Rock Islands of Palau.
We are currently developing a tool that expedites the classification and labeling of datasets from Island Conservation's project islands. We are also open to working with other organizations that can benefit from our technology (Examples include: poacher detection, rare animal tracking, population counting, etc).
At the moment, our solution is a post-processing tool for organizations that make use of off-the-shelf camera traps (Reconyx, Bushnell, etc). Our future plans include developing a low-cost camera that runs the classifier on an on-board microcontroller. Included in this design would be a propagation method using LoRaWAN, to transmit the detection event to a base station.