Drones, Internet of Things and Agriculture

Using 5G, Machine Learning and drones to increase agricultural efficiency

Comment

Inspiration

We wanted to create a project to improve efficiency and make people's lives better through technology. Agriculture was the perfect fit: drones, 5G and machine learning can be used to increase food production to levels not seen before. With these innovations, farmers can enjoy a higher standard of living and so can people all around the world. Improved food production will drop prices and take us closer to eliminating world hunger.

What did we learn?

That it's a lot harder than it seems at first glance! Although we had all programmed before, our team had no previous experience in machine learning. This meant we had to learn how to use the tools provided by Reply to train models and apply them to datasets. The drone simulation was also something new to us. Overall, picking an unfamiliar topic meant that we learnt a lot from this hackathon.

How was it made?

We used machine learning and neural networks to train and develop the models. We developed 2 models:

  1. Model to classify whether a corn leaf is healthy or infected. We used this data source to develop it: https://www.kaggle.com/qramkrishna/corn-leaf-infection-dataset
  2. Model to classify whether a seedling is maize or weed. We used this data source to develop it: https://www.kaggle.com/vbookshelf/v2-plant-seedlings-dataset

We used Figma to make prototypes of the desktop and mobile phone apps that allow the user to interface with the drone and access data onboard the drone and stored in the Edge Server and the Cloud.

On the drone simulation side, we used MAVSDK to control a simulated drone. QGroundControl was used to display the output on a map. There wasn't enough time for us to program the navigation completely, but by the end of the hackathon we could control the drone using Python and see its path on a map.

Challenges faced

This was a particularly challenging hack, since we essentially had to learn most things from scratch. The machine learning was particularly difficult, and we had quite a few issues trying to train the model. Drone navigation also turned out to be unexpectedly tricky.

What we made?

We made 2 AI Models which can be used to analyse drone footage and determine pest infestation as well as weed infestation.

Built With

Share this project:

Updates