GB-Salisbury: Aerial Imagery, Ground Truthing

Supervised Machine Learning for classification and prediction tasks are of importance to MOD and wider government.

Large advances in this area are not only due to access to large and various datasets but also to the known classes and attributes of the data.

These labelled classes are required to generate an error function which is used for optimisation purposes to learn and generate classifier models; generally the larger and more various the labelled dataset the higher the accuracy of the trained models.

The MOD generates and uses large datasets through sensing the battlefield.

Classification algorithms would provide benefit to the analysts by triaging large datasets into smaller priority datasets.

However, labelled, ground truthed data is sparse, yet it is this data which may provide the best advantages.

This requirement is for the generation of accurate labelling of aerial imagery datasets for use in further research, potentially through an online open challenge in using Supervised Machine Learning algorithms to data scientists.

The labelled datasets will allow advances in the area of Geographical Intelligence (GEOINT) using supervised machine learning methods.