Getting Predictions Back from the Edge

[MMarcial]

Getting Predictions Back from the Edge

Deploying a small battery operated MCU device to the edge is one thing. When the device is part of a larger ecosystem, getting the machine learning predictions and other data back to a central evaluation center is crucial.

This project uses

• the Edge Impulse Studio a machine learning model creation studio,
• a Sony Spresense MCU,
• an Seeed Studio moisture sensor with an analog voltage output, and
• the Arduino IDE.

Here a long range radio is investigated that uses

• the LoRa radio protocol,
• MCCI’s — Arduino LoRaWAN open-source library,
• The Things Network owned and operated by its users, and
• the Mathworks ThingsSpeak service.

The execution time for the moisture data to run thru the Edge Impulse DSP and Classifier is sub-millisecond!

The custom Edge Impulse model can be cloned here.

The Spresense code and overall project details are here.

[davidtischler_edgeim]

Wow this is awesome @MMarcial - great work putting this together!

[MMarcial]

Go Mobile with You Moisture Gauge

The original App released above was updated with physical location capabilities by exploiting the Sony Spresense onboard GNSS model.

The new program is here.

The architecture of the app was changed to allow one to adapt the code their purposes. For example, if you don’t want to use LoRaWAN, then delete files Spresense-LMiC.h and Spresense-LMiC.ino from the Arduino build folder.