Blog 4 - Refining Requirements

In week 9, we thought about how we can merge the two ideas, the Farm Data Analysis System and Digital Shepherd, into one. We decided it was most appropriate to focus our design around monitoring plants rather than moving livestock as they are more accessible for testing. Also, if we choose plants that are small in size, we do not need to spend a lot of money on buying hardware for large-scale monitoring. By creating a prototype for optimising plant growth on a small-scale, we can easily convert this to an industry application if we succeed.

We came up with a general set of requirements that is based around the raspberry pi grove kit and AI camera provided:

Must have:

• Control system that takes inputs from sensors of the environment in which plant is growing in.

• Process data.

• Allow the system to produce some response to input through at least 1 form of actuation.

• A simple interface to allow users to see sensor values.

Should have:

• A camera system that monitors plant growth

• A model verifying the health of the plant. (e.g. by analysing its leaves)

• Allow users to have some control over system settings.

• Have a more refined actuation system. (more forms of actuation: Fans, LEDs, Watering)

• Users can manually control the actuation of each subsystem.

• Feed new data back to the AI models to improve their accuracy (refine the model).

Could have:

• GUI for the user interface with plots of data over time.

• A model for plant growth predictions.

• Heating subsystem: heating element + thermometer (need to take safety precautions)

• Animal recognition model.

• A model to classify plants

In week 10, we refined these requirements so that we know specifically how to get started with the implementation and the various hardware we may need to buy in advance:

Refined Set of Requirements

Must have: Create a simple control system for plant growth.

1. Set up a system with three plants in a plastic container with devices connected.

   1. Design electronic circuits of the system.

   2. Plan out the hardware arrangement and necessary modifications to the container.

   3. Collect hardware needed and assemble them.

2. Get input from sensors in regular intervals and store them on the raspberry pi.

   1. Grove DHT11 humidity and temperature sensor (for air)

   2. Grove sunlight sensor

   3. Grove soil moisture sensor

   4. Temperature sensor (e.g. for soil)

3. Add lighting subsystem

   1. Solder/Assemble lighting subsystem control circuit

   2. Solder/Assemble the subsystem’s circuit (with PWM control)

   3. Attach LED strips to the container

   4. Attach control circuit

   5. Attach light sensor in the box to get most representative data

4. Add control to the system (simple PID control):

   1. If the air humidity is too high, turn on the fan

   2. If brightness is too low, increase the brightness of LED lights. if brightness is too high, decrease the brightness of LED lights.

   3. If ground humidity/ soil moisture is too low, add water.

5. Create a simple user interface.

   1. Create a simple console program that displays sensor readings and actuation status.

Should have: AI camera monitoring health, improved UI (graphs+Camera feed), More subsystems (watering subsystem, air circulation subsystem)

1. Add AI camera to the system

   1. Cut a hole for the camera lens so it can point directly at the plants

2. Set up a model in the Azure cloud, periodically upload data from Rpi.

   1. Train the model on data from the internet/manually created data

3. Perform data analysis on the sensor data

   1. AI model for inference on the data

      1. Create predictions

   2. The program allows input for optimal condition

4. Perform data analysis on the camera data

   1. Recognise each leaf

   2. Determine how dry each leaf is

   3. Detect abnormal properties on leaf e.g. spots

5. Add air circulation subsystem (fans+fan control system)

   1. Solder/Assemble air circulation subsystem control circuit

   2. Solder/Assemble the subsystem’s circuit (with PWM control)

   3. Cut out holes in the box for the fans and attach them to the box

   4. Attach the air humidity/temperature sensor in the box to get the most representative data

   5. Attach control circuit

6. Add watering subsystem

   1. Solder/Assemble the pump’s control circuit

   2. Solder/Assemble the subsystem’s circuit

   3. Attach water pump in the water container

   4. Attach liquid level sensor in the water container

   5. Add the water distribution system, either:

      1. Hydroponic: Add more water by directly pumping to the existing container

      2. Soil: either:

         1. Pump water through a pipe irrigation system

         2. Pump water to sprinklers above the plants

7. Continue AI model training with the system we have.

8. Research lighting subsystem

   1. Create 2 patches of the led control program.

   2. Compare the difference between simulated day and nights/constant lighting.

9. Allow the user control:

   1. GUI includes two modes: user control mode and automatic mode.

      1. Automatic mode = Users can enter the ideal temperature, humidity and soil moisture values for the subsystem to maintain in the GUI.

      2. User control mode = Users can turn on/off the fans, sprinkler and LED lights, and also adjust the brightness through the GUI.

Could have: Extra features - plant classification, stress level detection, heating subsystem, more advanced GUI, animal recognition.

1. Allow the camera to classify plants and detect stress levels of plants:

   1. Further training with more online pictures for different plants

   2. Need to research about detecting plant stress level

   3. Take in input from the camera to create predictions of how the plant might grow in the future.

2. Add the heating subsystem with safety precautions.

   1. Create a circuit with a heater and temperature sensor. The heater will turn on and off to control the temperature.

3. Advanced user interface with graphs showing sensor readings over time and diagrams to show predictions generated from the AI model.

4. Animal recognition:

   1. Train another model to allow the camera to classify them.