Machine Learning Based Crop Classification in Nepal

AI in Agriculture: In partnership with the UN World Food Program (UNWFP) in Nepal, Omdena’s AI changemakers built a deep learning model to fight hunger by locating, tracking, and improving the growth of crops such as rice and wheat. 32 collaborators from 18 countries took on the challenge of using machine learning for crop classification.

Nepal´s struggle with poverty

After a decade of unstable governments and armed conflicts, law and order are still tenuous in Nepal. One-third of the Nepalese population lives below the poverty line. Agriculture, the backbone of the Nepalese economy, is the main source of livelihood for 80% of the population. The potential of using AI and machine learning technologies in agriculture is on the rise. Therefore, the UN WFP and Omdena’s community took on the challenge of fighting hunger by building a machine learning based crop classification model. The solution can help to improve resource allocation for accelerating the growth of staple foods such as rice and wheat.

The solutions: AI in Agriculture to fight hunger

Right from the start of the challenge, the data collection team faced limited data access and low data quality.

Low-quality open-source data

Despite the obstacles, the collaborators built a crop identification model with 89 percent accuracy. Among many accomplished tasks, the community aggregated several data sources such as Sentinel satellite images of the Copernicus program and the EuroSAT dataset. In addition, the latest deep learning tools like super-resolution helped to improve the data quality further. As a result of the collaborative effort, a seven-point recommendation guide for creating satellite imagery datasets for agricultural applications was derived.

Details about the challenge and learnings can be found on our blog below.

The impact 

Where our machine learning model will be implemented

This challenge functioned as a pilot project for the UN WFP by using explicitly open-source data to prototype a solution. The challenge results built confidence that machine learning based crop classification can help governmental institutions to make data-driven decisions on where to allocate resources for crop growth most effectively.

Currently, we are defining a follow-up challenge to take the solution to the next level.

We are thanking all community collaborators for the amazing work done!