COFNet: A deep learning model to predict the specific surface area of covalent-organic frameworks using structural images and statistic features.

[Display omitted] • Prediction of experimental specific surface area of COFs based on structural images. • The proposed COFNet model integrates deep learning algorithms with attention mechanism. • The COFNet model innovatively accepts COF structural images and derived statistical features as hybrid inputs. • The hybrid inputs further enhance the prediction accuracy of the COFNet model. • The COFNet model outperforms available Zeo++ code in the prediction of COF specific surface areas. We present a new approach wherein the prediction of Brunauer-Emmett-Teller (BET) specific surface areas for Covalent-organic frameworks (COFs) is performed using a newly developed deep learning model (COFNet). This model integrates deep learning algorithms with an attention mechanism, and innovatively incorporating structural images of COFs and the statistical features derived from these images as model inputs. Results reveal that the COFNet can satisfactorily predict the specific surface area of COFs. It significantly outperforms the publicly available Zeo++ software. The developed COFNet model is a promising tool for efficiently predicting experimental BET specific surface areas of COFs. [ABSTRACT FROM AUTHOR]