Improving Food Security and Crop Yield in Senegal Through Machine Learning

Improving Food Security and Crop Yield in Senegal Through Machine Learning

A challenge initiated to help farmers know where to add water or fertilizer using data such as soil PH, temperature, and moisture levels, combined with other data sources. The data and predictions will also help to decide where to invest, and help strengthen the understanding of crop losses while maximizing revenues and minimizing losses.

 

The Global Partnership for Sustainable Development Data (GPSDD) seeks to leverage machine learning to help farmers cope with increasingly erratic weather, model the fastest route to markets and mobilities across livelihood zones, and detecting problems in fields with drones and others tools. GPSDD seeked to leverage machine learning to help farmers improve crop yield prediction.

 

The Problem

At a time where the world needs to produce more with fewer resources, AI could help to transform agriculture worldwide and especially in Senegal. The ability of agricultural equipment to help actors better think, predict, and advise farmers via a variety of AI applications presents Senegal with the potential to achieve food security in the country. Senegal is among those hardest hit by climate change, according to scientists, with populations that depend largely on agriculture losing their livelihoods due to worsening and recurrent floods and droughts.

Using technology, we can now, have better results to know what is going to happen and where in Senegal while promoting a data-driven agricultural system. Data such as soil PH, temperature, and moisture levels, combined with other data sources from ANSD and other stakeholders such as DAPSA, CSE, ANACIM, CSA, could be processed to show exactly when and where farmers should add water or fertilizer. The data will also help to decide where to invest, and help strengthen the understanding of crop losses while maximizing revenues and minimizing losses.

 

The Results

In these two months, Omdena’s collaborators were able to implement a Deep Learning model that predicts crop yield in Senegal following this schema:

Crop yield prediction - Source: Omdena

Crop yield prediction – Source: Omdena

 

The data used to get crop yield prediction using deep neural networks:

  • Remote sensing data downloaded with Google Earth Engine (GEE)
  • Ground truth crop yield data: we had yield data collected by IPAR for the production of maize, rice, and millet in 2014

Data engineering and pre-processing were used with the satellite imagery datasets they collected, before feeding them into Convolutional Neural Networks for a Transfer Learning process.

 

Transfer learning on satellite imagery - Source: Omdena

Transfer learning on satellite imagery – Source: Omdena

 

Then the team created an application using open-source satellite images to identify the crops and estimate the yields for any given area.

To read more about the whole process, check the articles below, where the collaborators shared the methodologies used and the essential code.

A demo of the application is attached below

 

Join our challenges here.

And find all our community benefits here.

Find more information on how an Omdena project works.

 

Applying Remote Sensing and Computer Vision for Farming Habitat Classification

Applying Remote Sensing and Computer Vision for Farming Habitat Classification

Origin Chain Networks (OCN) is a tech startup with the mission to forge a future of food we can trust. Their mission is to promote fairness and incentivise participation with a bottom-up, farmer-first data ownership model and an accessible mobile farming solution.

 

In this two-month Omdena Challenge, 50 AI changemakers built an open-source Earth Observation reference dataset for classifying commercial crops and peripheral habitats on-farm that can be used by food industry bodies to contextualize on-farm data that is self-reported by farmers.

 

The problem 

The collation of national datasets for farm-level environmental impacts is normally conducted at a governmental level and is usually determined using general calculations based on estimated performance. This information is never utilized by farmers in order to adapt or change for improved performance. Changing the flow of information from the bottom up will disrupt and dramatically improve the quality and accuracy of this information. This is reliant on engaging the farming community to deliver on required data. However, if this is achieved successfully there will be a novel and new market for the farm-level data. Farmers will be at the center of this data revolution and should see the benefits both in the supply chain and through government supports. 

By ameliorating, valorizing, and assuring the integrity of self-reported data against the public, the independent dataset we can solve the issue of trustless reporting, reputation management, and brand protection on the part of all actors. 

 

 

The project outcomes

The Origin Chain Networks *Agri-trust mobile farming service helps farmers to digitize and report compliance-based field data. Most recently, in the EU, with the advent of the Green Deal, compliance requirements encompass environmental measures as well as commercial and food safety objectives.

This challenge focus on land quality classifications: 

  • Commercial farming habitats including crops, grasslands, commercial forestry, large scale glasshouse, and polytunnel production as well as livestock-rearing in sheds (poultry, eggs, livestock) and horticulture under glasshouses and polytunnels.
  • Peripheral habitats including waterways and (non-commercial) habitats on farms. These may include uplands, wetlands, hedgerows, native woodlands, and leisure gardens.

 

The deliverables of the project are as follows:

  • Testing what (if any) useful crop and peripheral vegetation classifications can be derived from open-source satellite imagery.
  • Creating an annotated data set of commercial crops, grassland for livestock, and peripheral habitat classification
  • We require a data visualization methodology that allows for public access to the outcome where farmers and other stakeholders can view and provide feedback on the classification dataset. 

 

Ultimately, Origin Chain Networks can help farmers to improve strategic decision-making by enabling a broad overview and understanding of the impact of commercial and peripheral landscape practices, incentivizing and accelerating the adoption of positive environmental actions.

 

Join our challenges here.

And find all our community benefits here.

Find more information on how an Omdena project works

 

Detecting Weeds and Crops Using Drone Imagery

Detecting Weeds and Crops Using Drone Imagery

Skymaps is an agtech startup using remote sensing technologies and advanced image analysis for precision farming. Their mission is to improve practices in agricultural interventions and thus achieve sustainable agricultural production.

In this two-month Omdena Challenge, 50 technology changemakers built a computer vision model to identify weed species as well as crop types.  

 

The problem

Persistent herbicides can remain active in the environment for long periods of time, potentially causing soil and water contamination and adverse effects to non-target organisms. Herbicides can cause deleterious effects on organisms and human health, both by their direct and indirect action. In this project, you will build a solution to significantly reduce the usage of herbicides. 

 

The project outcomes

Building an advanced model that integrates into the Skymaps application. The project deliverables are as follows:

1. Detecting and identifying weeds and crops

The ML model is able to: 

  • Identify weed species based on previous annotations
  • Identify different crops (corn, cereal, sunflower, etc.)
  • Additional functionalities:
    • The user needs to be able to select the crop to “help” the model
    • The user selects the most probable / focus weeds to “help” the model

 

In addition, the project team annotated additional weed samples from the field. 

 

Weed annotation

Example: weed annotation

 

Impact of the solution: Less spraying of herbicides 

 

Weed detection

Source: Skymaps

 

The model output is a Geo-referenced vector file (shapefile) with the detected weed zones (polygons). 

 

2. Building an annotation tool to classify other features

An additional deliverable of the project is to develop an annotation tool for different samples to classify a variety of features (disease, crop damage, water, etc.). 

Data specifications 

All data is provided in this project. We use a combination of RGB and multispectral layers with resolutions of 5-30 mm/ px (ground sampling distance) and with Resolution 10-100 mm/px (Ground sampling distance).

 

Join our challenges here.

And find all our community benefits here.

Find more information on how an Omdena project works.