Crop Mapping and Field Boundary Detection

Agricultural Research Center (ARC)

Business Needs

Agriculture automation is a main concern and an emerging subject for every country to optimize providing and reserving the annual storage of food and to solve problems facing the agriculture sector, such as crop disease management, pest control, weed management, irrigation, and water management. Agricultural Research Center (ARC) wanted to proactively support the increase of vegetation production in different regions all over Egypt. The vision was to identify the various vegetation types/crops using AI and remote sensing to automate the recommendation of the appropriate crops suitable for each land types (e.g., moisture levels of soil or droughty). The customer also wanted a more efficient automated process to identify planted areas, and hence having a better way to predict the required amount of irrigation water and the expected crop yield. This also entailed a need for raising the efficiency of ARC staff by building their capacity to use the latest technology for rapid crop type identification and field boundary detection. ARC also wanted to enable farmer’s engagement through reporting their field observations.

ARC logo3

Solution Concept

The solution integrates satellite imagery, field observations, data feeds, and GeoAI analytical capabilities to support ARC in maximizing the various crops' yield from limited resources (agricultural land). This is achieved through field boundary detection and crop type detection and classification. The solution provides imagery analysis workflows that leverage Esri geospatial platform capabilities and utilize Azure AI products and services; namely: Azure GeoAI Data Science Virtual Machine (DSVM) and Azure Custom Vision Service.
The agricultural fields in Egypt are commonly distributed with relatively small sizes parcels that usually reduce the reliability of agricultural statistics in surveying cropland. Therefore, the identification of crop types will be implemented by using time series analysis of remote sensing data reflecting crop classification. This will support building an accurate crop inventory under complex landscape conditions based on the spectral characteristics differences of crops. For field boundary detection, a deep learning semantic segmentation model will be trained (using UNet) to detect agriculture field boundaries. This will be accomplished through an end-to-end capability for imagery AI workflows on top of Azure, using the Azure GeoAI DSVM.

The second is developing a scalable and flexible GIS platform that can be adopted across all companies, and finally building AGIBA team capacity (for both the staff to work on and the decision-makers to see what is going on) through providing extensive customized GIS training to different system users to work confidently and productively.

Golden Points

  • Crop detection enables more accurate yield estimation, and hence better monitoring of crop production and distribution.
  • Crop detection enables better estimation of irrigation water consumption, and hence better control of irrigation water loss.
  • Better understanding of the cultivated areas enables a more efficient crop production planning.

Used Software

  • ArcGIS Pro installed on a data science virtual machine (DSVM) image from Microsoft
  • ArcGIS Enterprise
  • Portal for ArcGIS
  • ArcGIS Pro
  • Spatial Analyst extension
  • Surevy123
  • Workforce for ArcGIS
  • Collector for ArcGIS
  • Operations Dashboard for ArcGIS