Seismic Velocity Anomalies Detection Based on a Modified U-Net Framework

Accurate and efficient reconstruction of hidden geological structures under the surface is main task high-resolution Velocity Model Building (VMB). The most commonly used methods in practice are Tomography Full Waveform Inversion (FWI), which rely heavily on initial model. Recently, deep learning types have received widespread attention performed well many tasks such as image segmentation classification. Therefore, it great significance to introduce algorithms into VMB procedure accelerate production cycle, especially for velocity anomalies detection, crucial a In this paper, modified U-Net framework proposed applied directly seismic shot gathers identify early stage VMB, can provide suitable guess following large-scale procedures FWI. numerical examples show power method synthetic data.