UC2 AUTOMATED PARCEL PRELIMINARY BOUNDARY DETECTION AND DELINEATION PROTOTYPE REVIEW

Aug 5, 2020

 

UC2 Prefilled application seeks to make it easier and more accurate for farmers to annually declare their lands by reducing the time needed to draw the new parcels boundaries or update the existing ones, thus, prevent the mistakes with input of the data on crops grown.

 

On July 15th during internal UC2 team webinar meeting Sinergise presented the prototype results of automated parcel preliminary boundary detection an delineation for pilot area of Lithuania base on Sentinel 2 data and two approaches: unsupervised machine learning and supervised deep learning.

 

After review and analysis of both results it was decided to use supervised deep learning method – ResUnet-a - as it is more accurate and generate parcel preliminary boundaries for all Lithuania territory in order to prepare for 2021 operational and timely testing. More details can be found in presentation presentation.