Anomaly Detection on ISPU

[guest_7]

Hello,

I am interested in implementing a static version of the NanoEdge AI Anomaly Detection for the ISM330ISNTR. That means I would like to us the trained knowledge from NanoEdge AI Studio for the library and not have to run a learning cycle after each power up, because my application requires many short uses, where a learning sequence each time would be very hindering. Is this at all possible?

Many thanks in advance :)

[Julian E.]

Hello @guest_7 ,

For the ISPU, no, it is not possible to skip the learning phase as the product was not designed for the use case you are describing...

ISPU doc: AI:NanoEdge AI Anomaly Detection library for ISPU - stm32mcu

For any MCU, you could though.

To do that, you need in the compilation step to check "Include knowledge from benchmark". In the zip file, you will have a new file (the knowledge.h).

In your application, you need to include the knowledge. Then you need to use this function:

enum neai_state neai_anomalydetection_knowledge(const float knowledge_buffer[]);

documentation: AI:NanoEdge AI Library for anomaly detection (AD) - stm32mcu

Have a good day,

Julian

[guest_7]

Thanks for the fast answer, but when selecting the ISM330ISN ISPU as the target MCU the "include knowledge from benchmark" is not an option. This seems to differ from normal Anomaly Detection projects on other MCUs, I guess due to the fact that the ISPU is directly on the sensor. Do you have other suggestions?

[Julian E. · Accepted solution]

Hello @guest_7,

You are right, my answer is wrong, I forgot that the ISPU is working differently than the rest of the NanoEdge AI Libraries.

Sadly, there is no solution to your issue because you are not supposed to use the ISPU like this. One of my colleagues contacted you in private to learn more about your project and see if there is not a better solution for you to use.

(I edited my last message no to spread wrong information)

Have a good day.

Julian