Emergency services increasingly use drones for life-saving missions. Now processing their data with neural networks is a new way to speed up response.
Climate change and COVID-19 have made conversations on how we respond to emergencies mainstream. We’ve become more aware of the essential workers who risk their lives to save lives or even just keep things ticking over. We’ve seen multiple applications of new technology fast tracked in the quest to address never-before-seen challenges.
In disaster relief, it’s great to see ever-widening applications for drones (also known as unmanned aerial vehicles or UAVs,) with their ability to give a fast and relatively inexpensive overview.
For example, during 2020’s locust invasion, Kenyan Red Cross used drones to assess impact on agriculture and livestock and the effectiveness of control efforts. In Oregon, California and other fire-ravaged US states, firefighters used drones to map fires and even to drop incendiary fireballs that help stop forest fires. These start small fires, removing fuel an approaching fire needs to spread.
Artificial brain takes to the sky
At Kaspersky, we’ve been working on a particular contribution to making it easier for authorities to respond to disasters and emergencies with drones and artificial intelligence (AI.) We’re piloting an emergency analytics system that uses a neural network. These are mathematical models or algorithms made up of interconnected nodes (neurons) working a little like a human brain.
Neural networks recognize objects by searching for matches in the same way we search our memory to recognize a face, a name or a word.
Drones gathering data and identifying objects
Of this technology’s many possible applications, the most important are in situations where human life is at risk. Drones collecting data for real-time analysis by neural networks can search for missing people, animals and houses washed away in floods. It can assess damage from aerial images after an incident or provide crowd management information, helping to avoid human crush incidents.
We’re working with drone manufacturers Albatros and have already installed the Kaspersky Neural Network in two of their drone models.
In developing the system, we focused on its capacity to detect small objects in limited visibility. For example, emergency services need to find buildings that survived after an earthquake and distinguish residential from non-residential buildings. In a flood, they need to know the amount of debris in a flooded area.
Machine learning technology like a neural network can process images and recognize objects instantly. In one of our tests, the Kaspersky Neural Network processed thousands of high-resolution images of rough terrain on a laptop with limited computing power in a few minutes.
Our neural networks self-teaches – each time you use it, results improve. Trained on tens of thousands of images, it can recognize and classify people, houses, cars and more. You can process data onboard a drone or on a server or laptop after the flight.
What makes the system unique is that it can process data onboard several drones in the air in real-time and display it for multiple operators in one web interface, meaning simplicity that could speed up response times.
Nikita Kalmykov, CEO of drone makers Albatros says, “The system is initially focused on incident analytics and people search. It can also help in the agricultural, construction and oil industries. It has the flexibility customers need for intelligent aerial surveys.”
Concerns around drones in emergency relief
Using drones that process and categorize information using a neural network comes with challenges. Drone technology researcher Faine Greenwood points out public distrust of drones has slowed their adoption in disaster relief. To help smooth over this turbulence, Humanitarian UAV Network educates people and develops standards for humanitarian drone use.
When training a neural network, as with all AI, it’s important to consider potential bias in data sets and work to reduce it. For example, can the system identify people of all skin colors, genders and body sizes? Can it identify buildings or plants seen in a wide range of countries?
Gathering data ethically and storing it securely is another area of concern. In their best practice guidance, Humanitarian UAV Network advises, “Only collect aerial data that has a clear value for specific humanitarian purposes. Other data could be used in discriminatory or other negative ways, or exploited by third parties to harm local communities.” The International Red Cross also issues a handbook for data protection in humanitarian action.
One thing is clear: In a world of growing challenges in preserving human life, every new tool is worth having. While AI rapidly transforms every part of our lives, drones using neural networks have an important role in disaster relief. It’s no surprise humanitarian organizations and emergency services increasingly use drones in their work. Processing information with neural networks can add another level of efficiency and effectiveness to their important work. I hope we’ll see this technology widely adopted, and with it, many lives saved.