Crop For Change!

Team Friesenberger

Kumar Ayush and Kalpesh Krishna

IIT Bombay

What's the idea?

• Every piece of land in a village prefers certain crops over the others.
• Farmers tend to specialize in growing a certain crop over others, over years of experience.
• Why can't farmers pool in their resources, plant crops in the most optimal manner, and share profits?
• We aim at predicting an optimal distribution using Deep Learning.

Deep Learning Model

• All code was written in TensorFlow.
• The network is a feed-forward neural network with one hidden layer having 1000 neurons.
• Training is done using AdamOptimizer and exponentially decaying learning rate.
• 16 softmax layers have been added as the final layer, each operating on a part of logits.
• Average cross entropy has been used for training.
• Trained for 25 epochs on generated data with batch size of 100. (Demo)

Data Generation

• An inherent distribution of soil yield is generated.
• A distribution of available seeds is generated randomly.
• Greedy algorithm used to generate the data.
• Algorithm on next slide.

Visualization

• Our webapp has 2 major components.
• Crowd Sourcing - The first interface allows users to enter farming data. This is supposed to append the training data. 

Visualization

• Our webapp has 2 major components.
• Optimal Yield - This file executes a TensorFlow evaluation script and prints out the distribution for each plot of land.
• There is an interface to choose input parameters to be fed to the evaluation script. (Demo)

Visualization

What's Left?

Other Applications

Thank You!