Smart Agro Probe- blog 4: Building the tinyML model

27 May 2021

The accelerometer data can be used to detect the minute vibrations during the elephant's walk or poachers vehicle or illegal deforestation

etc.I would like to create 4 different models here to detect different scenarios.

Fullscreen contentimage_201526.html Download

<html><head><title>Jive SBS</title></head>
<body><font face="arial,helvetica,sans-serif">
<b>Error</b><br><font size="-1">
An general error occurred while processing your request.
</font></font></body></html>

CLI tool showing that the device is connected to the edge impulse and you can start training the model.

Fullscreen contentimage_201527.html Download

<html><head><title>Jive SBS</title></head>
<body><font face="arial,helvetica,sans-serif">
<b>Error</b><br><font size="-1">
An general error occurred while processing your request.
</font></font></body></html>

Web UI of edge impulse training data

Fullscreen contentimage_201528.html Download

<html><head><title>Jive SBS</title></head>
<body><font face="arial,helvetica,sans-serif">
<b>Error</b><br><font size="-1">
An general error occurred while processing your request.
</font></font></body></html>

Web UI of creating a model with the trained data

Fullscreen contentimage_201529.html Download

<html><head><title>Jive SBS</title></head>
<body><font face="arial,helvetica,sans-serif">
<b>Error</b><br><font size="-1">
An general error occurred while processing your request.
</font></font></body></html>

Feature explorer

Initial training results

Fullscreen contentimage_201531.html Download

<html><head><title>Jive SBS</title></head>
<body><font face="arial,helvetica,sans-serif">
<b>Error</b><br><font size="-1">
An general error occurred while processing your request.
</font></font></body></html>

Download the library

Import the library by going to arduino >> sketch >> include library

Complete tutorial by edge impulse is here.

You'll need to add the raw feature data before uploading the code to the device. Use the static buffer code and add the following code before your main library definition

This is needed for the SAMD21 boards, if you get any error use the documentation here

#include <cstdarg>
#define __STATIC_FORCEINLINE                   __attribute__((always_inline)) static inline
#define __SSAT(ARG1, ARG2) \
__extension__ \
({                          \
  int32_t __RES, __ARG1 = (ARG1); \
  __ASM volatile ("ssat %0, %1, %2" : "=r" (__RES) :  "I" (ARG2), "r" (__ARG1) : "cc" ); \
  __RES; \
 })

Results from the model detecting the standing still motion with 98% accuracy.