SOLUTIONS

Crop Detection

The challenge

Due to the importance of crop recognition and field monitoring in making national food and groundwater policies and economic plans, it is crucial to crop identification. Remote sensing has been widely used as a useful tool for crop identification at a large scale due to its high temporal resolution, wide coverage, and low cost. However, cleaning and extracting accurate and useful crop information from these images is challenging, time consuming, and requires a lot of manual work. Moreover, fields in Saudi Arabia are small compared to other countries, posing a challenge to crop recognition methods. However, by using the right techniques of artificial intelligence, machine learning, and image processing, crop identification is automated and more accurate. This saves time, improves accuracy, enables better production management, and reduces environmental impacts.

What we did

By integrating multispectral satellite imagery from low-resolution Landsat8 data to a higher resolution multiband WorldView imagery, and by developing methodologies that with a high precision detect new crop fields and accurately classify different crop types, we provided our clients with accurate data that are critical for improving food security and agricultural growth. In the process, we used spectral signatures and multitemporal data to identify crop types. Moreover, we used many vegetation indices to improve accuracy (e.g., leaf area, soil type, NDVI, and LAI). We also used pre-existing crop types and sites that were manually detected to validate our methodology, detect new sites, create accurate polygons for crop fields, and quantify the total area of crops.

Our Impact

We provided our clients with a model that enabled them to monitor agricultural activity from satellite imagery more quickly and at a lower cost. Our model also accurately classified crop types and predicted yields. We analyzed each crop type more closely using spectral signatures, temporal growth trends, and vegetation indices highlighting crop information that varies from one crop to another, thus classifying and comparing crops.