Choose timezone
Your profile timezone:
Powered by Indico
v3.3.5
Live in YouTube: https://www.youtube.com/watch?v=Fp75uNtRQ-0
We will start this last webinar with an overview of the AI4EOSC platform, its main features and a brief introduction on how users can take advantage of it. The objective is to provide a global picture of the platform in order to present some of the functionalities that will be explained in the following talks by real users.
This talk will focus on deep learning based thunderstorm nowcasting suited for agricultural applications. It will provide an overview of meteorological background as well as describe the experience with the AI4EOSC platform.
Using AI4EOSC’s federated learning and composite AI solutions, we are pushing the boundaries of disease detection models in agriculture, which will be integrated into existing national advisory platforms, such as eDWIN, operated by AI4EOSC partners (WODR and PSNC), and accessible to farmers, advisors, and scientific institutes. We offer individual risks of cumulative risk calculations for the most common crops and related diseases including potato blight and Cercospora in beet. This presentation will show a practical use of the provided feature of a integrated plant protection use case.
The third AI4EOSC project use case revolves around unmanned aircraft system (UAS)-based thermal imagery and how this form of data can be leveraged to find thermal points of interest in our cities' environments. This spans from detecting thermal bridges on building rooftops for energy retrofits, over identifying all manner of common urban hot-spots, to segmenting thermal anomalies for pipeline leak detection.
Three modules have been implemented on the platform that enable the training of object detection and semantic segmentation models as well as prediction with well-known CNN- and transformer-based architectures. Not only are the platform's resources leveraged to perform both these tasks but other functionalities are also made use of, such as MLFlow for experiment tracking.
We finish this webinar with an external use case of the AI4EOSC platform about satellite imaging. The objective of the use case is to develop a data-driven fusion method that combines the data from Sentinel-2 and Sentinel-3 satellite missions to obtain new products with the best of both in terms of spatial and spectral resolution (images with 21 spectral bands at a spatial resolution of 10m). For this purpose, a Super Resolution Generative Adversarial Network (SRGAN) is trained, and its performance is compared with the one of a Super Resolution Convolutional Neural Network (SRCNN), as well as with those of some traditional data fusion algorithms, such as pansharpening and bicubic interpolation. AI4EOSC platform resources are leveraged for both code development and training.