Bench Talk

Edge Impulse Use Case: Emergency Preparedness and Wildfire Detection

(Source: neillockhart - stock.adobe.com)

Machine learning has the potential to fundamentally change the way we exist and operate within our world. At its core, machine learning is just a mathematical way to predict future events based on previous data. However, when applied to big challenges, machine learning can pave the way for a safer and brighter future for humanity.

Machine learning holds great promise in the field of emergency preparedness. Here, researchers are beginning to investigate ways to leverage this technology to predict emergency events, such as extreme weather situations, with enough notice to avoid them altogether or at least reduce their impact.

Machine learning offers a unique opportunity for wildfire detection, especially when implemented on the edge. In this blog, we’ll discuss the promise of machine learning for wildfire detection, why the technology needs to be deployed on the edge, and how we were able to use Edge Impulse to create a proof of concept for this technology.

Up to this point, one of the most pressing issues facing humanity has been how to adapt and react to seemingly spontaneous natural events such as hurricanes, earthquakes, and wildfires. In general, we’ve yet to find a way to anticipate these events, and even if we do, we often only realize once it’s too late to protect ourselves.

This is especially true for wildfires: We are aware of the conditions that are conducive to wildfires, but their actual onset is often considered a random event. As a result, we’ve been unable to protect ourselves from the spread and damage of wildfires, resulting in lost human and animal lives as well as destroyed woodlands. Today, with the rise of global warming and climate change, predicting and preparing for wildfires is seemingly more important than ever before.

Researchers have begun looking at ways to use machine learning technology to address this issue.

Scientists have long known that certain variables, such as the temperature and humidity of air and soil, are key indicators of an area’s susceptibility to wildfires. However, putting together mathematical models that can take these variables and accurately predict the occurrence of wildfires has been challenging.

With machine learning, this has all changed. By taking data from key indicators and feeding them into a machine-learning model, wildfires can be predicted with a high degree of accuracy. With this knowledge, we can prepare for wildfires by evacuating animals and removing flammable biomass from an area, ultimately reducing the damage and severity of wildfires.

In most cases, machine-learning applications are expected to run on the cloud, where big servers provide the processing power needed to perform machine-learning computation. However, in the instance of wildfire detection, this computation needs to be moved to the edge for several key reasons.

In a wildfire detection application, a device will be deployed consisting of a number of environmental sensors, such as humidity and temperature, and machine-learning algorithms are then run on these collected data. In this scheme, two options exist: Send the data to the cloud for processing or process them on the edge.

A major challenge with cloud processing in this context is the remote deployment of these devices. These devices are generally deployed in remote locations, like in the middle of a forest, where wildfires might actually take place. In these isolated locations, network connectivity is very limited, making it difficult, if not impossible, to communicate all of the sensor data to the cloud for processing. Instead, with machine learning on the edge, all of the data and processing can be kept on the local device. The only thing to be communicated with the outside world would be a warning in the rare instance that a wildfire was determined to be likely.

Another benefit of edge computing is that it may require less power expenditure. In the vast majority of cases, a remotely deployed sensing device will be powered by small lithium-ion or lithium-polymer batteries. In these cases, replacing the battery is not a realistic option; hence, the device is only useful as long as its battery is alive.

In general, one of the biggest power consumers for an Internet of Things device is the power spent wirelessly communicating with other devices in a network. In a cloud computing scheme, the device will burn significant power simply by communicating the large volume of sensor data to the cloud, ultimately shortening battery life.

Instead, with edge computing, less wireless communication means less power expenditure. Edge computing enables sensors with long battery life and hence a greater opportunity to provide emergency preparedness information.

In the course of developing our wildfire detection proof-of-concept device, we encountered a variety of significant challenges and solved them with Edge Impulse.

One of these major challenges pertained to sensor fusion. In our device, we take data from a number of disparate sensors, including sensors for the temperature and humidity of both air and soil, and we try to understand these different data streams. Doing this requires sensor fusion, which is the process of merging data from multiple sensors for a more encompassing view of the environment.

In general, implementing sensor fusion is a difficult task full of many unique complexities and algorithms. Luckily, Edge Impulse offers a built-in suite of tools meant specifically for facilitating and implementing sensor fusion on edge devices. With this tool, we were able to successfully, and rather easily, design a system that captures, aggregates, and formats our data so that the data can be fed into our machine-learning model.

Furthermore, Edge Impulse made it easy for us to select and train our model as well as to deploy it to our microprocessor. In our case, our microprocessor was an ATSAMD51-based core on a Seeed Studio Wio Terminal.

Needless to say, our project would not have been possible if it weren’t for the tools and resources provided by Edge Impulse.

For all of history, humanity has been forced to react to natural events and emergencies as they occurred. Now, with the advent of machine learning, we finally have the ability to predict and prepare for emergencies in ways that were previously unfathomable.

Wildfire detection is an important cause that is becoming increasingly paramount, but, due to the unique restrictions of the application, it requires edge computing. Thanks to Edge Impulse and the tools and resources it provides, we were able to develop a proof-of-concept wildfire detection device that can accurately alarm park rangers and other officials if a wildfire is imminent.

At the end of the day, this technology has the potential to save the lives of humans and animals as well as to prevent the destruction of our fragile woodlands.