Applied AI & AI in business /

Picture-based crop insurance for small-scale farmers around the world

April 5/4 min read
  • Anish JainWinner of the No-Code AI Challenge

Smallholder farmers are increasingly exposed to weather extremes but lack access to affordable insurance products to protect their livelihoods from catastrophic crop damage.

Picture-based insurance (PBI) improves crop insurance for small-scale farmers around the world, where images from a smartphone camera keep a record of a crop’s growth and record any damage events that will affect insurance payouts. PBI is a great way for insurers to verify events and monitor crop growth.

02/ The use case

In India, where I am from, agriculture is a billion-dollar industry. There are about 100 million farmers and if only 10% of those (around 10 million) would buy this service for their crop insurance for $20 a month, this would result in $250.000 in revenue just from one country, one month. So, in one year that would result in $30 million in revenue. Since the cost of training the AI model would add up to about $10.000 to $20.000 a year, the profit margin looks very promising.

In this project, I automate one part of the data processing pipeline which is estimating the growth stage of a wheat crop based on an image uploaded by the farmers. My model takes in an image and outputs a prediction for the growth stage of the crop sown, on a scale from 1 (crop just showing) to 7 (mature crop).

03/ The dataset

I am using a wheat crop dataset, consisting of images of varying length and width, but I resize them to 224*224. I am using a dataset from this website, a great platform for research and learning about agriculture.

04/ Building the model

I choose to build my AI model using Peltarion. One of the great things about the platform is that it's easy to use if your data is prepared. Before this, I used the Google Cloud platform which is good but not as beginner-friendly. I used the pre-trained restnet152 model.

When training the model, the root mean square error was 0.43 for the best model in a test dataset. My model reached an accuracy of 76% on the test dataset - it definitely needs more training

Model performance

05/ Deployment

I used Bubble for the deployment as it's so easy to design things and connect with APIs.

A user has to upload the picture of their farmland and the model then tells us at which growth stage the crop is at. Based on this, we decide to give them their insurance money or not.

How the application works:

  1. Farmer downloads our web or phone app
  2. Within the app, they enroll as many land sites they prefer
  3. For every enrolled site, they upload an initial overview picture
  4. Every few days from sowing to harvest, farmers upload new pictures of the same site with the same exact view frame aided by the app
  5. Our model is used to predict the potential land damage from the pictures
  6. Farmers who suffered crop damage received insurance payout directly into their bank account

06/ Final remarks

I believe that this is just the beginning of a super cool product. I discussed this with my father who is doing some farming, and he really thinks it has great potential. My grandparents were part of a village where agriculture played a major role in their community, at times the only source of income. A platform like this, with this kind of model and application, could help farmers a great deal and drive the economy forward.

  • Anish Jain

    Winner of the No-Code AI Challenge

    Anish was the winner in the "Best Business Innovation" category in our No-Code AI Challenge. The competition took place in February 2021.

Applied AI & AI in business topics

    More to read