The CoeusAI QGIS plugin is designed for exploration of multiband geospatial datasets. It lets the user iteratively train and retrain segmentation models in seconds. A combination of Deep learning and traditional machine learning is used, leveraging the best of both methods.
Our QGIS plugin, CoeusAI, is designed for exploration of multiband geospatial datasets. It lets the user iteratively train and retrain classification models in seconds. Traditional machine learning approaches can lack intelligent predictions, while deep neural nets typically require vast amounts of training data and compute. Our tool combines a pretrained UNet for feature extraction with a RandomForest for classification. We therefore leverage strong points of both. The use of a UNet makes it possible to take into account complex patterns and encode these in a feature space. The RandomForest is then trained on top of those features with little labeled data, and only seconds (or minutes) of compute time required. This allows for an interactive and explorative user experience.
CoeusAI uses the functionalities of the command-line-tool and Python package, Pycoeus, it's main dependency, and makes them available with a user-friendly graphical user-interface to use in QGIS.
Processing archaeological sensor data recorded with drones using machine learning
Pycoeus is a command-line tool and python package designed for exploration of multiband geospatial datasets. It lets the user iteratively train and retrain segmentation models in seconds. A combination of Deep learning and traditional machine learning is used, leveraging the best of both methods.
