Optimization of artificial neural networks models applied to the identification of images of asteroids’ resonant arguments

The asteroidal main belt is crossed by a web of mean motion and secular resonances that occur when there commensurability between fundamental frequencies the asteroids planets. Traditionally, these objects were identified visual inspection time evolution their resonant argument, which combination orbital elements asteroid perturbing planet(s). Since population affected is, in some cases, order several thousand, this has become taxing task for human observer. Recent works used convolutional neural network (CNN) models to perform such automatically. In work, we compare outcome with those most advanced publicly available CNN architectures, like VGG, Inception, ResNet. performance first tested optimized overfitting issues, using validation sets series regularization techniques data augmentation, dropout, batch normalization. three best-performing then predict labels larger testing databases containing thousands images. VGG model, without regularizations, proved be efficient method large datasets. Vera C. Rubin observatory likely discover up four million new next few years, use might quite valuable identify populations minor bodies.