利用Encoder-Decoder框架的深度学习网络实现绕射波分离及成像

利用单纯绕射波场实现地下地质异常体的识别具有坚实的理论基础，对应的实施方法得到了广泛研究，且有效地应用于实际勘探。但现有技术在微小尺度异常体成像方面收效甚微，相关研究多数以射线传播理论为基础，对于影响绕射波分离成像精度的因素分析并不完备。相较于反射波，由于存在不连续构造而产生的绕射波能量微弱并且相互干涉，同时环境干扰使得绕射波进一步湮没。因此，更高精度的波场分离及单独成像是现阶段基于绕射波超高分辨率处理、解释的重点研究方向。为此，首先针对地球物理勘探中地质异常体的准确定位，以携带高分辨率信息的绕射波为研究对象，系统分析在不同尺度、不同物性参数的异常体情况下绕射波的能量大小及形态特征，掌握绕射波与其他类型波叠加的具体形式；然后根据相应特征性质提出基于深度学习技术的绕射波分离成像方法，即利用Encoder-Decoder框架的空洞卷积网络捕获绕射波场特征，从而实现绕射波分离，基于速度连续性原则构建单纯绕射波场的偏移速度模型并完成最终成像。数据测试表明，该方法最终可满足微小地质异常体高精度识别的需求。

Diffraction wave separation and imaging with deep learning network based on Encoder-Decoder framework

Identification of underground geological abnormal bodies only by diffracted wave fields has a solid theoretical foundation, and corresponding implementation methods have been widely studied and effectively applied in actual exploration. However, the existing techniques have made slight progress in micro-scale abnormal body ima-ging, and the most of related studies are based on the theory of ray propagation, which have an incomplete analysis of factors affecting the accuracy of separated imaging. Compared with reflected waves, diffracted waves ge-nerated by discontinuous structures have weak energy and usually interfere with each other, and environmental interferences further eliminate the diffracted waves. Therefore, high-precision wave field separation and indepen-dent imaging are the key research directions of current interpretation processing based on the ultra-high resolution of diffracted waves. First, according to the accurate positioning of micro-scale target bodies in geophysical prospecting, this paper takes the diffracted waves with high-resolution information as the research object and systema-tically analyzes the energy magnitude and morphological characteristics of the diffracted waves under abnormal bodies with different scales. The goal of this process is to master the specific superposition forms of diffracted waves and other types of waves. Second, from the corresponding characteristic properties, the paper proposes a separate method of diffracted waves based on deep learning technology. New method introduces the Encoder-Decoder framework and atrous convolution to capture the properties of diffraction wave. In virtue of the outputted unique diffraced wave field, we could calculate the migration velocity via continuing velocity criterion. Finally, migrated data and velocity model are obtained. Data tests show that the proposed method can realize high-precision identification of micro-scale geological anomaly bodies in industrial production.