Group Actions
Engagement
  • Author Author: dougw
  • Date Created: Date Created
  • Views 1626 views
  • Likes 15 likes
  • Comments 7 comments
Related
Recommended

Save the Bees - Machine Learning

dougw

Intro
The project kit has not arrived yet, but I want to keep the project moving with another interesting post. This blog covers recognition of various bugs by a computer vision system that uses machine learning to identify the various bugs. It also introduces the interface PCB, 3 of which will be used - one for each system node.

image

Bug Recognition
This video shows the HuskyLens in action recognizing 2 different bees, a spider, a cicada, a grass hopper and a ladybug.

HuskyLens

I tried to zoom in on the HuskyLens display....it isn't as dark as the camera implies...

In case you are wondering how much work is involved in training the HuskyLens to recognize insects, the next video is a quick demo of how the HuskyLens is trained to recognize multiple objects.

Machine Learning Simplified

Although I find the buttons a bit awkward, the designers of HuskyLens did a lot of work to make this powerful little system simple to train and use. In case you are wondering about the technology - the Huskylens uses a Kendryte K210 dual core RISC-V processor.

System Review

The Arduino MKR WAN 1300 will be used as the controller in each of 3 nodes in the system. To do this, a PCB was designed that allows interfacing all the different peripherals in each system. All three nodes will use the same PCB with a MKR WAN 1300, but each node will have different peripherals and will have different firmware. Actually one of the nodes will also be built in 2 different configurations, one will feature a HuskyLens and the other a Nicla Vision module.

image

Interface PCB Design
The PCB can connect an Arduino MKR WAN 1300 to 3 different cameras, 3 different LCD displays, 2 different speech synthesizer modules and an audio amplifier.
Here is the interface schematic:

image

The PCB has been laid out and ordered and will be described further when it arrives.

Discussion
Some progress has been made on machine training, insect recognition and PCB design. However the project feels like it is off the rails. The build period is scheduled to run from Jan 4 to March 17 and the project kit has not been received yet. There is a lot of work left to be completed with less than a month remaining. It is great to see the HuskyLens vision system in action. I am impressed with its power.

Next steps
In the next blog the HuskyLens will be connected to an Arduino MKR WAN 1300 to allow it to use names for each bug instead of generic labels.
When the new speech module arrives, the Arduino MKR WAN 1300 will be able to respond to Husky recognized bugs and give each critter its own voice.
When the project kit arrives the work of communicating what the edge system has discovered will begin.

Links:

 Two Bee Or Not Two Bee 

 Save the Bees - Machine Learning 

 Name That Bee 

 Bee The PCBee 

 Show Me The Bee! 

 The Lora & Husky Show 

 Image Conversion to Integer Array for LCD Display 

 LoRa Transmission of Images 

 Save The Bees Wrap Up - Blog 9 

 Save The Bees - Design Challenge 

Parents

  • Congrats on finding the HuskyLens. I tend to get stuck on trying to reinvent things from scratch. The Huskylens seems to be a big step up from the PixyCam and less expensive. How does it perform real time ? Do you anticipate the same time invested on your part in NiclaVision to load & train the ML  as with the HuskyLens ?  

Comment

  • Congrats on finding the HuskyLens. I tend to get stuck on trying to reinvent things from scratch. The Huskylens seems to be a big step up from the PixyCam and less expensive. How does it perform real time ? Do you anticipate the same time invested on your part in NiclaVision to load & train the ML  as with the HuskyLens ?  

Children
  • in reply to robogary

    I expect NiclaVision to be a much steeper learning curve for me. HuskyLens has made the process really simple with automated dataset collection and no programming required to do object recognition/classification. There will still be programming involved to make use of the recognition results, but that is unavoidable. So far I am impressed with real time performance, it is perfect for this application.

  • in reply to robogary

    One objective is to consider the skill sets of potential users - in this case bee keepers and farmers. The technology needs to be simple enough for anyone to handle without requiring them to learn a whole other discipline from their chosen field. The main reason for including the HuskyLens is just in case the NiclaVision is too complex. It also sets a goal for where NiclaVision development can go.