🌱 plantnotplant 🍂

Plant classification subsystem

Hacking towards an automous AI rover hardware and software platform classify plants!

Research plan:

1. Deep learning pipeline for state transitions

• Recognize center of plant and perform classificaiton
• Recognize end of track without relying on hardware

2. Deep learning pipeline for data labeling

• From hand-written label, recognize species of plant
• From hand-written label, apply data labels
  • Height of plant
  • Number of plants
  • ??

3. Deep learning classification

• Collect N camera positions at a single plant
• Is 3D CNN better, or just a single position?

Botony:

Ocimum × africanum, (Thai Basil)

Origanum vulgare (Oregano)

Hardware:

2. Rasperry-pi as master

• Camera: Night / Standard / Fisheye?
• Python stack
• Keras/Tensorflow
• Open CV

1. 2-axis mounted camera controlled by Arduino

• Stepper motor for linear
• Servo for camera scanning

Motor 1.7 A / stage

DRV 8825 max continuous current: 1.5 A

Target current: 1.3 A (derated)

VREF = Current limit / 2 = 0.65V

SEE: https://forum.arduino.cc/index.php?topic=415724.0

Communication protocols:

Arduino slave, connected by I2C bus to RPi master.

Bluetooth PS3 controller.

WiFi networking.

Software:

Arduino motor control software link

Onboard systems: MuleAI

Tensorflow 1.8.0

Numpy 1.14.4

Offline training: corral

3D CNN

---

Table of Contents

• Features
• Prerequisites
• Quickstart
• Environment variables
• Code style
• Testing
• New Version
• License

---

Features

TODO

Prerequisites

TODO

Quickstart

...

Environment variables

TODO

...

Code style

. ​

Testing

Automatic tests are setup via Travis, executing tox. Our test use pytest framework.

New Version

The bumpversion.sh script helps to bump the project version. You can execute the script using as first argument {major|minor|patch} to bump accordingly the version.

License

todo