Uploaded February 2026 | Updated September 2026, 5 hours ago
This is the project video from Daniel Radovanovic's bachelor's thesis on his Sunspots Observation Platform. It was established in 2025 at the Bern University of Applied Sciences in Switzerland.
The bachelor's thesis describes the development of a web-based platform for the Rudolf Wolf Society for the Digitization and Automation of Previously Manually Conducted Observations of Solar Activity. The society has been conducting systematic solar observations for many years and thus stands in a long tradition of visual sunspot observation.
Background and Objectives
The traditional observation process requires the manual recording of sunspots on daily logs using a telescope, a clipboard, and a pen. This analog method is time-consuming and prone to observational errors. The aim of this thesis was to develop a modern, web-based system that digitizes the observation process and supports it through machine learning.
Solution
Within the scope of this thesis, a complete web application with a PostgreSQL database, FastAPI backend, and Next.js frontend was implemented. At its core is an image-processing pipeline for SDO-HMI satellite imagery. A series of image-processing steps identifies relevant sunspot regions, which are then rectified to compensate for perspective distortion at the solar limb. A convolutional neural network (YOLOv8) was trained to automatically detect and classify sunspot groups using the modified Zurich classification. The implemented human-in-the-loop system enables users to correct model predictions, thereby continuously expanding the training dataset and leading to steady improvements in detection accuracy.
Result and added value: The developed system offers the Rudolf Wolf Society significant added value. The platform simplifies and accelerates the observation process, reduces sources of error, and enables the automated analysis of solar images. The web application integrates user and instrument management, observation documentation, and a comprehensive machine-learning training system into a unified interface.
This is the project video from Daniel Radovanovic's bachelor's thesis on his Sunspots Observation Platform. It was established in 2025 at the Bern University of Applied Sciences in Switzerland.
The bachelor's thesis describes the development of a web-based platform for the Rudolf Wolf Society for the Digitization and Automation of Previously Manually Conducted Observations of Solar Activity. The society has been conducting systematic solar observations for many years and thus stands in a long tradition of visual sunspot observation.
Background and Objectives
The traditional observation process requires the manual recording of sunspots on daily logs using a telescope, a clipboard, and a pen. This analog method is time-consuming and prone to observational errors. The aim of this thesis was to develop a modern, web-based system that digitizes the observation process and supports it through machine learning.
Solution
Within the scope of this thesis, a complete web application with a PostgreSQL database, FastAPI backend, and Next.js frontend was implemented. At its core is an image-processing pipeline for SDO-HMI satellite imagery. A series of image-processing steps identifies relevant sunspot regions, which are then rectified to compensate for perspective distortion at the solar limb. A convolutional neural network (YOLOv8) was trained to automatically detect and classify sunspot groups using the modified Zurich classification. The implemented human-in-the-loop system enables users to correct model predictions, thereby continuously expanding the training dataset and leading to steady improvements in detection accuracy.
Result and added value: The developed system offers the Rudolf Wolf Society significant added value. The platform simplifies and accelerates the observation process, reduces sources of error, and enables the automated analysis of solar images. The web application integrates user and instrument management, observation documentation, and a comprehensive machine-learning training system into a unified interface.
