Research On EfficientNet Algorithm For Morphological Classification Of Galaxies

The morphology of galaxies is related to the formation and evolution of galaxies,and structural properties provide valuable information on the formation and evolution of galaxies and are important for understanding the past,present and future universe.Therefore,its morphological classification is an important link in the follow-up research of galactic astronomy.However,with the appearance of massive astronomical observation data,the automatic analysis method of astronomical data has been paid more and more attention.The advanced deep learning EfficientNetV2 backbone network uses more than 240,000 photometry images from Galaxy Zoo 2 as initial data for experimental tests,focusing on the automatic classification algorithm of galaxy morphology and its improvement.The main research contents are as follows:(1)Screening and preprocessing of galaxy morphology images.The Galaxy image data used in this thesis is Galaxy Zoo 2,an online crowdsourcing project that asked participants to describe galaxy morphology based on color images,in which more than 16 million morphological classifications were derived from about 300,000 galaxy data released by the SDSS.In this thesis,we first explain the selection criteria of galaxy categories provided in the second Galaxy Zoo,and then screen out clean samples of 6 categories according to the criteria.Then,by means of data enhancement,a small number of category samples are expanded to eliminate the influence of sample imbalance on the robustness of the model.Finally,the sample was preprocessed by size dithering,flip,rotation,color distortion and other operations to enhance the generalization ability of the model.(2)The deep learning algorithm EfficientNetV2 is applied to the classification of galaxy morphology.After comparison experiments with the same series of classical and cutting-edge deep learning algorithm models AlexNet,ResNet,MobileNetV2 and RegNet,the eight models EfficientNet and the three models EfficientNetV2 are shown as the galaxy image classification experiments.The experimental results show that the EfficientNetV2-S has better characterisation capability for the data set.(3)The improved EfficientNetV2-S model and proposed the EfficientNetV2-S-Triplet7 algorithm.Based on the advanced deep learning backbone network EfficientNetV2,an improved model named EfficientNetV2-S-Triplet7 was constructed to realize the automatic classification of galaxy morphology,analyzing the different types of attention mechanisms and the effects of nodes used on the network performance.The experimental results show that the thesis EfficientNetV2-S-Triplet7 algorithm achieves the best results in classification Accuracy,recall ratio and F1-score.In 9375 test images,the three index values can reach 89.03%,90.21% and 89.93% respectively,and the Accuracy rate can reach89.69%,ranking the third among other models.The results show that the proposed method can be effectively applied to the task of morphological classification of large-scale galaxy data.