Searches for low-surface-brightness galaxies (LSBGs) in galaxy surveys are plagued by the presence of a large number of artifacts (e.g., objects blended in the diffuse light from stars and galaxies, Galactic cirrus, star-forming regions in the arms of spiral galaxies, etc.) that have to be rejected through time consuming visual inspection. In future surveys, which are expected to collect hundreds of petabytes of data and detect billions of objects, such an approach will not be feasible. We investigate the use of convolutional neural networks (CNNs) for the problem of separating LSBGs from artifacts in survey images. We take advantage of the fact that we have available a large number of labeled LSBGs and artifacts from the Dark Energy Survey, that we use to train, validate, and test a CNN model. That model, which we call DeepShadows, achieves a test accuracy of 92.0%, a significant improvement relative to feature-based machine learning models. We also study the ability to use transfer learning to adapt this model to classify objects from the deeper Hyper-Suprime-Cam survey, and we show that after the model is retrained on a very small sample from the new survey, it can reach an accuracy of 87.6%. These results demonstrate that CNNs offer a very promising path in the quest to study the low-surface-brightness universe.

在银河系调查中搜索低表面亮度星系（LSBG）时，会遇到大量伪影的困扰（例如，来自恒星和星系，银河卷云，恒星臂区中形成恒星的区域的混合在天体中的物体）螺旋星系等），必须通过耗时的目视检查才能将其排除。在未来的调查中，预计将收集数百PB的数据并检测数十亿个对象，这种方法将不可行。我们调查了使用卷积神经网络（CNN）解决从调查图像中的伪影分离LSBG的问题。我们利用以下事实：我们从“暗能量调查”中获得了大量标记的LSBG和伪影，可用于训练，验证和测试CNN模型。这个模型，我们称之为DeepShadows达到92.0％的测试准确性，相对于基于特征的机器学习模型而言，这是一个显着的提高。我们还研究了使用转移学习来适应此模型以对来自更深层次的Hyper-Suprime-Cam调查的对象进行分类的能力，并且我们表明，在对来自新调查的非常小的样本进行训练后，该模型可以达到一定的准确性占87.6％。这些结果表明，CNN为研究低表面亮度的宇宙提供了非常有希望的途径。