%0 Journal Article %A Kumar, Amarjeet %A Gary, Hampson %A Rayment, Tom %T Adaptive subtraction using a convolutional neural network %D 2021 %J First Break, %V 39 %N 9 %P 35-45 %@ 1365-2397 %R https://doi.org/10.3997/1365-2397.fb2021066 %I European Association of Geoscientists & Engineers, %X Abstract Adaptive subtraction is a well-established routine technique in seismic data processing that relies on being able to match, in some sense, one dataset to another. It is a necessary part of many processing steps such as de-noising and de-multiple. The same matching methods are also used for advanced imaging steps such as non-stationary estimation of the Hessian as part of image-domain least-squares migration. Predictions of noise are often not as accurate as we would like because there are often shortcomings in the prediction method and because pragmatic acquisition compromises are made. The ubiquity of adaptive subtraction in the seismic processing workflow means there is always motivation to improve its performance. In this work, we propose a deep-learning approach for the adaptive subtraction of predicted multiple models from the input recorded data. It turns out that it is straightforward to use deep-learning networks to perform adaptive subtraction and as a result, benefit from the power of complex non-linear filters. Inspired by the idea of a convolutional autoencoder, we have developed a supervised convolutional autoencoder for adaptive subtraction. We demonstrate the proposed approach on synthetic and field data examples and compare the results using conventional adaptive subtraction methods to illustrate the effectiveness of our approach. %U https://www.earthdoc.org/content/journals/10.3997/1365-2397.fb2021066