标题： SiameseLSRTM：使用孪生网络增强最小二乘逆时偏移 显示英文标题

标题： SiameseLSRTM: Enhancing least-squares reverse time migration with a Siamese network

摘要： 最小二乘逆时偏移（LSRTM）是一种基于反演理论的成像方法，通过优化理论迭代更新反射率模型以最小化观测数据和模拟数据之间的差异。 然而，在实际数据应用中，基于Born近似的模拟数据，比如基于简化物理假设如声波假设，通常无法充分反映观测数据中的复杂性。 因此，我们开发了SiameseLSRTM，这是一种新颖的方法，采用由两个具有共享权重的相同卷积神经网络（CNN）组成的暹罗网络来测量模拟数据和观测数据之间的差异。 具体而言，暹罗网络中的共享权重CNN能够从观测数据和模拟数据中提取可比较的特征，促进更有效的数据匹配，从而最终提高成像精度。 SiameseLSRTM是一个自监督框架，在迭代的LSRTM过程中更新网络参数，无需大量的标记数据和长时间训练。 我们使用两个合成数据集和一个来自陆地地震勘探的实际数据集评估SiameseLSRTM，结果显示它产生的成像结果分辨率更高且更准确，相比传统的LSRTM有所改进。 显示英文摘要

摘要： Least-squares reverse time migration (LSRTM) is an inversion-based imaging method rooted in optimization theory, which iteratively updates the reflectivity model to minimize the difference between observed and simulated data. However, in real data applications, the Born-based simulated data, based on simplified physics, like the acoustic assumption, often under represent the complexity within observed data. Thus, we develop SiameseLSRTM, a novel approach that employs a Siamese network consisting of two identical convolutional neural networks (CNNs) with shared weights to measure the difference between simulated and observed data. Specifically, the shared-weight CNNs in the Siamese network enable the extraction of comparable features from both observed and simulated data, facilitating more effective data matching and ultimately improving imaging accuracy. SiameseLSRTM is a self-supervised framework in which the network parameters are updated during the iterative LSRTM process, without requiring extensive labeled data and prolonged training. We evaluate SiameseLSRTM using two synthetic datasets and one field dataset from a land seismic survey, showing that it produces higher-resolution and more accurate imaging results compared to traditional LSRTM.