Cultivate the World

Cultivate the World Logo

Inspiration

Food production is what drives the development of the world. In many developing countries, the crop yield isn't sufficient to satisfy the needs of an ever-growing population. We decided to pursue the difficult task of solving this issue, and created Cultivate the World.

What it Does

Our system involves a network of low-cost, portable weather stations, which are easy to manufacture and distribute throughout the planet. These weather stations harvest data about the environment, including parameters such as temperature, humidity, soil moisture, air pressure, wind speed, and more. The data is then sent to a central sever, and then processed using machine learning techniques in order to predict useful weather patterns and warn farmers of impending weather formations that may cause crop failures.

How We Build It

The raw data is packaged using a Raspberry Pi Zero, a small and efficient computer. Data packets are then sent over an existing radio network, such as GSM, or our own network, like APRS with QPSK. We leverage cloud computing to centralize this enormous amount of data, and process it effectively. We use an LSTM (long-short term memory) recurrent neural network with dynamic weight matrices (implemented using genetic algorithms) to properly interpret the data and detect relevant patterns. Depending on the network coverage of the area, we would either use WiFi, GSM, or APRS to communicate the weather analysis back to the farmer.

Challenges We Ran Into

A lot of the problems we encountered were caused by hardware. For example, we ended up spend a lot of time looking for components like a 5V regulator. We were eventually able to solve this problem by taking apart several power banks and several others. We also had issues with computers crashing because of CAD (computer aided design). Additionally, we faced one main challenge with integrating our successful machine learning algorithm with our graphical user interface through the use of Firebase. While these challenges delayed our timeline, we were able to make up for most the lost time by working more efficiently and resourcefully.

Accomplishments

At the end, we were able to assemble a working prototype of a weather station. Costing less than $25 and roughly the size of two decks of cards, our weather station could collect data and feed it to a primitive LSTM recurrent neural network.

What We Learned

Throughout our 24 hours of non-stop hacking, we learned about the power of resourcefulness and setting team deadlines. While this hack definitely required some new skills, we mostly benefited from the opportunity to apply knowledge and theory. Specifically, we applied machine learning, something relatively new to all of us.

What's Next

We hope to carry Cultivate the World into the future by implementing more powerful machine learning techniques, gathering more training data, establishing a dependable communications system for station to server data transfer. Most importantly, we see the necessity for Cultivate the World in developing countries.