A team of US enthusiasts uses microphones and ML to identify wildfires

Compiled from 3dprint

To aid in early detection of wildfires, a team of technicians from nonprofit education provider New Collar Network enlisted the help of Knowles in audio and smart sensing to provide an audio-monitored smart sensor in a 3D printed housing that uses machine learning (ML) to differentiate between random forest sounds and fire sounds to alert authorities of a possible fire.

Filip Perez, Jed Beddo and Alec Kerr from the New Collar Network team received the Knowles AISonic Hardware Grant.

Using Knowles' advanced hardware, the device can identify fire ignition, reignition, and arcing of power lines to indicate possible fire risk. In addition, any one device can pass signals to other devices, such as field cameras, to provide forest service firefighters with instant fire information.

To demonstrate proof-of-concept and prototypes of early wildfire detection systems, the team will use Knowles' AISonic IA8201 Raspberry Pi development kit. This all-in-one package brings voice, audio edge processing, and machine learning recognition capabilities to devices and systems for a range of new applications. The latest development kit, launched in September 2021, bundles features for testing, prototyping, and debugging voice and audio capabilities as well as integration into smart homes, consumer technology, industrial, and more.

Knowles designed the new kit to allow product designers to prototype innovations in an easy and fast way to address emerging use cases. The development tools accelerate the development of contextual voice perception, ML acquisition, and real-time audio processing during the design process.

Because the Knowles development kit is so small, the team says it could easily be ported directly into a product to prevent wildfires. Additionally, they decided to 3D print a housing for the device so it can be easily attached to a tree trunk. Alternatively, they could 3D print other housings, such as one with magnets that could be used on metal surfaces. Finally, as an add-on, the team hopes to make the "IA8201 an always-on sensor unit" by powering it with solar power.

New Collar Network early wildfire detection sensor.

The team consists of Filip Perez, Jed Beddo and Alec Kerr. The Fab Lab Hub is a member of America Makes and a hands-on training center for digital fabrication skills.

“We would initially promote this for first responders or park rangers to demonstrate how well the product and hardware performs in outdoor conditions. Eventually, there will be a civilian version for people living in dry, fire-prone areas; and a campsite version that would allow campers to always know where the nearest wildfire is in relation to their site, giving advance warning and even monitoring of campsites.”

Early detection projects like this can save more forest land and keep people safe there, and reduce fire response times through more accurate location details. In addition, thanks to 3D printing technology, products can move from prototype to production faster.

The team was one of five winners of the Knowles AI Competition, along with a team from Malawi looking to build low-cost, small wind turbines using locally available materials and a Spanish team looking to solve the problem of sleep apnea.