The project aims at the development of algorithms for cross-calibration of hyperscout with S-2 and the development of the phi-sat 1 brain for cloud detection.
The team is assessing Machine Learning methodology, techniques, and a suite of applications (ranging from Radiometric calibration of a single band L2A data, Calibration of data products, e.g. NDVI or LAI and Radiometric calibration of a hyperspectral spectrum L2A data). The results from PhiSat-1 mission operations proved that is possible to detect clouds during day time using a neural network on HyperScout 2 Visible Near InfraRed spectral data. Additional activities have been identified in order to complete the cloud detection capabilities including also night time using HyperScout Thermal Infrared data.
The project extension aims to:
- Verify whether is possible to classify clouds in-orbit during night-time using the thermal infrared channel of HyperScout 2
- Developed synthetic HyperScout 2 TIR nocturnal imagery to be used for training, and asses whether the quality of that data directly affects the performance of the Network.
- Further develop the NightView neural network according to the HyperScout 2 needs
- Check whether cloud detection at night can be performed employing deep learning with thermal infrared as produced by HyperScout 2. Currently, the existing convolutional neural network is trained using the MODIS sensor on-board the Aqua and Terra satellites and the associated cloud mask.