Improving Digital Advisory Services for Rural Farmers Using Predictive Analytics and Satellite Imagery

This Omdena Local Chapter Challenge runs for 4 weeks and is a unique experience to try and grow your skills in a collaborative and safe environment with a diverse mix of people from all over the world. 

You will work on solving a local problem, initiated by Omdena Cracow, Poland Chapter.

The problem

We have seen traction in demand for rural digital advisory services, however current systems for digital advisory are focused on the broad delivery of extension services based on a large number of farmers. AI can revolutionize extension services through the provision of individualized advisory based on several data elements (on-farm data, satellite imagery, remote sensing, and GIS) thereby increasing the value of extension services to the individual farmer. Although use cases are being built in other development agencies and countries, we have not seen greater traction on AI and other technology integration in IFAD-supported projects. This could be an opportunity to develop a Proof-of-Concept (POC) and develop a potential use case for scale.

The goals

1. Getting LandSat/Sentinel to work at 10 meters. 

We aim to explore the satellite API and get a better resolution. Having a spatial resolution of 20 meters might work, but it makes the images appear really zoomed out. To configure the pull-in/out, we need key code and API examples.

2. Creating a mask for field boundary delineation.

As soon as we receive the images at the correct spatial resolution, we will follow the national GIS paper that was submitted. According to the paper, clustering images must be 4000 pixels by 4000 pixels, which is the spatial resolution required for clustering. Using the clustered images, we can create a mask.

3. Making dynamic time wrapping. 

Once we have the field boundaries, with the Tensorflow autoencoder we can do dynamic time warping (compare the sequences of series of crop growth to non-crops). 

4. Training a model for crop recognition. 

If we have a mask with decent accuracy, we can train the model to recognize crops. Once we are able to identify a crop, we can collect and scrape agricultural data and use multi-feature classification models with an output feature. We can train multi-classification models to predict crop yield, crop growth, etc.

5. Developing low-cost commercial products. 

The farmers can input their GPS location and the number of hectares they are farming and a predicted yield and fertilizer recommendation will be generated as a result. The application can be enhanced with additional features.

Why join? The uniqueness of Omdena Local Chapter Challenges

Omdena Local Chapter Challenges are not a competition or hackathon but a real-world project that will grow your experience to a new level.

A unique learning experience with the potential to make an impact through the outcome of the project. You will go through an entire data science project lifecycle. This covers problem scoping, data collection, and preparation, as well as modeling for deployment.

And the best part is that you will join the global and collaborative community of Omdena with tons of benefits to accelerate your career.

Read more on how Omdena´s Local Chapters work