Give a man a fish and you feed him for a day. Teach a man to fish and you feed him for a lifetime. It is very obvious that, if we want to minimize world hunger, one has to stop unnecessary food loss due to preventable diseases. We also strongly believe the app will help farmers reducing their CO2 emission and the use of pesticides and fungicides by using the app proactively.

What it does

Our trained algorithm compares the images taken from suspicious plants to both healthy and infected ones. Afterwards it recommends the user about how to deal with the issues. It is especially made for local Ghanians’ who might not have access to the Internet or the advantages our high-tech society has.

How we built it

Most of the time we separated tasks around our group. While Paul started off by cloning data and training the first model, Philemon made the front-end of the application meet the mockup Tanja designed. Jan swapped between different positions in our team provided all the research data needed. When the first model was ready to be implemented over tensorflow into the javascript environment, the other half of our team focused more on the tools Microsoft provided us. The final solution for our Cocoa model was to apply image transformation in Azure ML to the basic data, store the edited ones in the Storage Explorer and to train our model in The last hours of our hackathon were used to implement all the different kinds of codes into a react-native environment and watch the sunrise on Sunday.

Challenges we ran into

One big challenge was to train the algorithm in a way that it has a good balance between a high accuracy and still not being overfitted. We had two datasets. Using the first we had a very precision but in the end a very low impact on both environment and economy. However using only the second one may gave us a beneficial effect but with less accuracy. That’s why we decided to use both sets to make our model suitable for more farmers and at the same time generated more training data using Nevertheless the most challenging part was making the different models fit and together under our own benchmark of accessibility.

Accomplishments that we're proud of

Our application is usable for a very diverse Ghanian community. We even went some steps further by not only providing helpful information about the diseases themselves in Twi but also speech and non-text based hands-on advices for farmers to prevent their yields from future impacts. Moreover our application is usable.

What we learned

Not everything will work out the way you expect it. The first hours we definitely underestimated the challenge and ran into some huge implementation models. We’d definitely split our time a little bit differently and focus on a minimum value product earlier.

What's next

We are already in discussion with enactus-Ghana to get them on board to help us bring the application to the farms and for local support & insights.

Built With

Share this project: