Project Description

Crop Damage Detection

Project scope

Traditional scouting methods were always used by corn farmers in determining the extend of crop damages in case of extreme weather events. These kinds of methods are tedious and difficult to ascertain the extent of the damage. ArcGIS, Machine Learning and drone mapping provides a faster, efficient and accurate way to evaluate these damages.

The objectives of this project include: Examining the cornfield and identifying damaged corn areas, estimating the area of the unused cornfield and approximating the average corn height and plant density.

Data Sets

The geodata of the cornfield of approximately 9.43 hectare was acquired using a drone. The data obtained was in form of point cloud and images. The data is registered similarly to a basemap with WGS 1984 Web Mercator projection reference system and vertical units in meter, hence creating a georeferenced integrated spatial dataset.

Challenges & solutions

The acquired geodata were comprised of multiple 4-band images and point cloud – LAS (with same geographic relation) datasets. The images were then mosaiced to form an orthomosaic raster image. Prior to the data analysis, radiometric corrections were done to correct for the measured brightness value of pixels, band to band error, and geometric and panoramic distortions that occurred during the acquisition process.

To enable for photointerpretation, the orthomosaic raster image was processed. To allow for identification of different features on the digital image, an image classification algorithm was done. Both the supervised and unsupervised classification were performed on the orthomosaic image , hence identifying damaged corn areas and estimating the unused areas on the field. Using a fully trained algorithm, the individual pixels of the orthomosaic were classified with respect to their spectral properties.

The LAS dataset was used to generate geospatial products from the raster image e.g. extraction of the Digital Elevation Model (DEM) and Digital Surface Model (DSM). These are variables needed for band arithmetic in deriving corn heights. These are variables needed for band arithmetic in deriving corn heights. For estimation of corn growth stage, precision and accuracy in photointerpretation of the cornfield, an estimation of the NDVI (Normalized Difference Vegetation Index) was also done on the raster images.

Project outcome

An approximate of 5.51% of the 9.43 hectares of the cornfield area was identified as bare soils, while around 0.69% was detected to be the part of the cornfield where the corn was damaged. At the time the data was obtained, the corn was at its reproductive growth stage.

Also an average corn height of 2.1 meters was derived. The average plant density was estimated at 29,600 plants per hectare. It could also be deduced that the maize was in the reproductive growth phase at the time of uptake.

Category

GEO AI
Computational Lifescience
Spatial temporal Analysis
Computer Vision

Technologies

Image segmentation
Support Vector Classifier

Download

Social Sharing

Other Use Cases in this category

All Use Cases at a Glance

Contact

Stefanie Supper
CEO

Contact

Stefanie Supper
CEO