Abstract Algorithms for automated processing of seismic records are being constantly upgraded, and the tasks of data analysis are becoming more complex. Most algorithms require preliminary preparation of the data. This preprocessing is either very simple, such as frequency filtering, or highly sophisticated to extract specific properties of the signal. Adequate preprocessing can increase the efficiency of the further analysis by order and more. However, specialized preprocessing cannot be used for solving other tasks or with other postprocessing algorithms. We consider the solutions that do not result in the significant loss of information and that can be used for solving any tasks. The main goals of the preprocessing are to reduce the noise level, to remove the anthropogenic noise, and to reduce the dimensionality of the data. We assume that deep neural networks of certain architecture are used for further data processing; however, this does not preclude from the application of other algorithms. As the preprocessing of seismic data, in this paper we consider wavelet transform, autoencoder, and some other algorithms.

中文翻译：

---

用于深度神经网络分析的地震数据预处理的可能性

摘要 地震记录的自动化处理算法不断升级，数据分析的任务也越来越复杂。大多数算法需要对数据进行初步准备。这种预处理要么非常简单，例如频率滤波，要么非常复杂以提取信号的特定属性。充分的预处理可以提高按顺序等进一步分析的效率。但是，专门的预处理不能用于解决其他任务或与其他后处理算法一起使用。我们考虑不会导致大量信息丢失且可用于解决任何任务的解决方案。预处理的主要目标是降低噪声水平，去除人为噪声，并降低数据的维数。我们假设特定架构的深度神经网络用于进一步的数据处理；然而，这并不排除其他算法的应用。作为地震数据的预处理，本文考虑了小波变换、自编码器和其他一些算法。