Output behavior prediction of Tm/Ho:YAP integrated lasers via ResNet-MLP multimodal fusion regression

This study addresses the challenge that traditional models cannot accurately describe the output performance of Tm/Ho:YAP integrated crystals under strong thermo-optical coupling. We propose a dual-branch deep regression framework that combines residual convolutional neural networks (ResNet) with multilayer perceptrons (MLP). By jointly analyzing crystal temperature distribution images and key experimental parameters, the method captures laser dynamics under complex thermal loads with high precision. Experimental results show that the model significantly outperforms existing approaches in output power prediction, achieving a prediction error as low as 0.58% during dual-wavelength operation, thereby providing a new, to the best of our knowledge, perspective for the modeling and optimization of thermo-optically coupled solid-state lasers.