Signal processing is key to embedded Machine Learning

When we hear about machine learning - whether it's about machines learning to play Go, or computers generating plausible human language - we often think about deep learning. Lots of unstructured data gets thrown in a complex neural network with billions of parameters, and after a very expensive training stage the model learns the task at hand.

This is a companion discussion topic for the original entry at

Hello thanks for the great job. Am doing a project on Manhole detection using IoT and machine learning… The aim is always to build a system that can send timely updates inform of sms and emails to the responsible personnel.

Usually the best way to use IoT to detect and monitor manhole statuses in cities is by sending signals from sensors like the proximity sensor to a microcontroller which converts them into useful data by using the ADC in it. And through the code on it, it generates out puts data that is now sent to the GSM or GPRS and then messages and emails can be sent to the technical personnel alerting them about the Manhole cover status ie. Open or closed.

Is it possible for me to omit the microcontroller from this circuit?

How possible is it for me to get raw data from the sensors and I directly send it to the machine learning model without having to first go through the microcontroller as it’s commonly done?

So that this data from sensors is directly sent to a well trained machine learning model, so that the model generates updates of whether the Manhole is open or closed basing on its prior training threshold values.

This would help me reduce the size of the circuit by omitting having a board consisting of a microcontroller having a code, that runs to give outputs, so that all this decision making of whether the Manhole status is open or closed will be decide by the machine learning model.

Kindly advise… If it’s possible let it be quick.

Electrical engineering student in Uganda east Africa.

Hi @Wilsonmk so if you omit your MCU then where would you run the ML model? The ML model needs to run on something that can execute software, so the MCU is the logical place for that.

@janjongboom, is it possible to review the images of this article? It is not open on my machine. I tried with 2 different browsers.

Clear your history and cache memory of your browser and then try it again. Hopefully, it works but if not, you can drop your email, and I will send it to your email.