A lot of “AI at the edge” talk is about vision – and that’s quite natural because there are a lot of interesting applications in that space. To make classification decisions on video or high resolution images you need the processing power of a good GPU, or perhaps an applications processor with a dedicated AI processor, such as the NXP i.MX8M Plus – and we have a module which does that. Or, you can send the data to the cloud, and do the AI processing there.
But for every camera at the IoT edge, there are hundreds of simpler sensors, measuring temperature, pressure, movement, monitoring motor voltages and currents, and so on. Usually connected locally to a simple microcontroller, which has the job of interfacing to whatever communications protocol is needed to transfer the data stream.
The pervading problem with this approach is the need to constantly transmit a reasonably high volume of data back to the cloud. This will probably need a 4G modem, meaning much higher power consumption than would otherwise be needed. Plus it’s not without cost penalty to have this sort of constantly on connection, both in terms of network fees, and security management.
The ideal solution would be to perform the AI-based classification task locally.
Because of the high demands of image based machine learning and AI, there’s a tendency to overestimate the processing power needed for this sort of task. It turns out that a microcontroller based on a low-end ARM Cortex-M series core – of the sort that might already be locally available alongside the sensor – can easily perform this task, using as little as 10KB image size. You need a specially parameterized machine-learning model to achieve this efficiency. But you don’t need to be any sort of expert.
The SensiML toolkit – which you can try out without charge – does the clever AutoML stuff for you, and also helps to collect, label and present the data to the model, as well as providing tools to assist in testing and perfecting the model.
Have a look here for more info: https://www.directinsight.co.uk/sensiml-analytics-toolkit/
Pricing is here: https://www.directinsight.co.uk/sensiml-subscription-plans/