Bench Talk

(Source: Amy Corrine Richards, Conservation X Labs)

Machine learning (ML) is a revolutionary technology that has facilitated unprecedented applications. Still, in many ways, the technology is limited by the fact that it is often confined to data centers and high-performance computers. Artificial intelligence (AI) is an inherent part of ML, and bringing it to the edge is challenging this notion and unlocking new use cases that were unattainable until now.

One such use case is wildlife conservation. Organizations that work in the wildlife conservation space are performing ML inference on the edge to study, track, and protect endangered species. Tools such as tracking cameras and environmental sensors are producing data that can be coupled with ML to better inform conservation and protection efforts.

In this post, we examine how Conservation X Labs worked closely with Edge Impulse, the leading development platform for ML on edge devices, to jointly develop solutions for bringing edge AI to the wildlife conservation field.

Conservation X Labs aims to develop solutions powered by innovation and technology to prevent the sixth mass extinction. They focus on the drivers of the crisis, not the symptoms, to confront the issue at its source.

To support these efforts, Conservation X Labs engineered innovative monitoring and tracking technologies, such as Sentinel, to help prevent wildlife trafficking, stop the spread of invasive species, and contribute to healthy ecosystems. A Google Coral–based AI toolkit, Sentinel connects to devices like trail cameras to give them edge AI capabilities. For example, with Sentinel, Conservation X Labs can upgrade a standard trail camera to a motion-tracking camera capable of leveraging ML to perform automatic wildlife detection and classification for conservation studies.

In the design of these kinds of systems, one of the most important capabilities is real-time monitoring. With real-time monitoring, a trail camera can observe, detect, and capture live images or videos of animals. The key to successful real-time monitoring is low-latency computation.

Generally, ML applications rely on cloud-based data centers to execute compute-intensive ML algorithms. However, for applications like Conservation X Labs’, cloud computing is not a feasible solution.

A major reason why cloud computing is not suitable for Conservation X Labs’ application is that wildlife detection and tracking devices are often deployed in isolated and remote locations. Such locations can make it very difficult for monitoring devices to find access to the high-bandwidth wireless connectivity required to meet cloud computing requirements. Out in the field, in a place like the jungle, cellular connectivity is virtually nonexistent.

Further, wireless communications may require higher energy and security costs. For wildlife conservation devices to most effectively do their job, they need the longest battery life possible, as battery replacements for remotely deployed cameras are an unrealistic option. Hence, the goal is often to limit the amount of data sent back and forth over the system.

All these factors led Conservation X Labs to one conclusion: The deployed tools need to leverage edge computing. By running algorithms on the edge, Conservation X Labs designed wildlife-monitoring tools that offer real-time performance without the cost and time overheads that come with cloud computing.

Edge Computing Challenges

Conservation X Labs quickly encountered significant challenges in designing edge-AI devices for wildlife preservation.

One challenge was the diversity of models required to track different kinds of animals, as each animal needs its own unique dataset and model training. The number of cloud resources required to continually train new models was enormous, and the associated financial cost was equally exorbitant, ranging in the thousands—an unsustainable amount from a business perspective.

Another major challenge was keeping up with the dynamic and fast-paced ML field, particularly its current state-of-the-art technologies, including advanced algorithms, new libraries, and evolving dependencies. To execute the conservation efforts most effectively, Conservation X Labs applied the most cutting-edge tools possible—not an easy task when the state-of-the-art changes frequently. Keeping up with the field is not only difficult but also undesirable, as it pulls developers away from focusing on potentially more important matters, like the solution effectiveness.

The Role of Edge Impulse

After trying many other tools, Conservation X Labs discovered Edge Impulse as an excellent solution to these challenges.

Edge Impulse’s platform makes the development, optimization, and deployment of ML models at the edge extremely easy and accessible. The platform enables developers to manage the workload at a high level, encompassing everything from data preparation and data selection to model selection, model training, and model deployment, including device-specific binaries.

Edge Impulse completely automates these processes and uses the most up-to-date libraries and dependencies to ensure that the ML solutions are based on the cutting edge. In turn, developers can deprioritize these more meticulous backend tasks.

To protect biodiversity and nature crucial to life on this planet,Conservation X Labs needed advanced technologies that improve the speed and scale of conservation efforts. Today, ML on the edge has enabled unprecedented use cases unlocking a clear path forward for addressing the looming extinction crisis.

Thanks to Edge Impulse, Conservation X Labs developed edge devices to monitor, detect, and ultimately protect endangered wildlife. Conservation X Labs believes these advanced technologies can help restore balance to the natural world and prevent future crises from occurring.