Browsing by Author "Pöntinen, Mikko"

One of the main factors currently limiting geophysical and geological studies of asteroids is the lack of visual and near-infrared (Vis-NIR) spectra. European Space Agency’s upcoming Euclid mission will observe up to 150,000 asteroids and gather a large amount of spectral data of them in the Vis-NIR wavelength range. Asteroids will appear as faint streaks in the images. In order to exploit the spectra, the asteroids have to first be found in the massive amounts of data to be obtained by Euclid. In this work we tested two methods for detecting asteroid streaks in simulated Euclid images. The first method is StreakDet, a software originally developed to detect streaks caused by space debris. We optimized the parameters of StreakDet, and developed a comprehensive analysis software that can visualize and give statistics of the StreakDet results. StreakDet was tested by feeding 4096×4136 pixel images to the software, which then returned the coordinates of the asteroids found. The second method is machine learning. We programmed a deep neural network, which was then trained to distinguish between asteroid images and non-asteroid images. Smaller images were used for this binary classification task, but we also developed a sliding window method for analyzing larger images with the neural network. After optimizing the program parameters, StreakDet was able to detect approximately 60% of asteroids with apparent magnitude V < 22.5. StreakDet worked better for long streaks, up to 125 pixels (corresponding to an asteroid with a sky motion of 80 "/h) while streaks shorter than 15 pixels (10 "/h) were typically not found. The neural network was able to classify the brightest (20 < V < 21) streaks with up to 98% accuracy when using very small images. When analyzing larger images, the sliding window algorithm produced heat maps as output, from which the asteroids could easily be spotted. The machine learning algorithm utilized was fairly simple, so even better results may be obtained with more advanced algorithms.