A Preliminary Study on Crop Classification with Unsupervised Algorithms for Time Series on Images with Olive Trees and Cereal Crops

Abstract

Satellite imagery has been consolidated as an accurate option to monitor or classify crops. This is due to the continuous increase in spatial-temporal resolution and the availability of free access to this kind of services. In order to generate crop type maps (a valuable preprocessing step to most remote agriculture monitoring application), time series are built from remote sensing images, and supervised techniques are widely used to classify them. However, one of the main drawbacks of these methods is the lack of labelled data sets to carry out the training process. Unsupervised classification has been less frequently used in this research field. The paper presents an experimental study comparing traditional clustering algorithms (with different dissimilarity measures) for the classification of olive trees and cereal crops from time series remote sensing data. The results obtained provide crucial information for developing novel and more accurate crop mapping algorithms.