Fig. 8

(a) FD-Conv model demonstrates outstanding performance in testing with new \(\mathrm Mg_2B_2O_5\)+\(\mathrm MgB_4O_7\) foam ceramic microstructure images, achieving the highest R@1 value and F1 score among all models, indicating its efficient recognition capability and low false positive rate in predicting the microstructure of materials. (b) The experiment by adjusting the depth of the Blocks shows that increasing the depth can improve the prediction performance and feature extraction ability of the model, but it is necessary to balance the feature extraction depth and the generalization ability of the model to avoid overfitting.