The identification of partially resolved space objects in an image and matching it to a database of known objects is useful for many scenarios. Machine learning can identify objects at various distances, relative attitudes, and phase angles. In this study, a convolutional neural network is constructed and evaluated. Performance as a function of phase angle is used to evaluate the capabilities of the neural network. The network’s ability to correctly identify objects blurred to various degrees is also assessed. A method for novelty detection is explored. Numerical results suggest that the use of deep learning may be a viable option for the identification of partially resolved objects.

中文翻译：

---

利用卷积神经网络识别空间图像中部分分辨的物体

在许多情况下，识别图像中部分解析的空间对象并将其与已知对象的数据库匹配非常有用。机器学习可以识别各种距离，相对姿态和相角的物体。在这项研究中，构建并评估了卷积神经网络。性能作为相角的函数用于评估神经网络的功能。还评估了网络正确识别各种程度模糊的对象的能力。探索一种新颖性检测方法。数值结果表明，深度学习的使用可能是识别部分解析对象的可行选择。