Taking into account the challenges posed by tritium due to its radioactive contamination and cost, we propose designing a simulation program to calibrates the fusion yield between materials by transfer learning. We use the generated DD target data to carry out transfer learning on the initial neural network, so that the initial predictive DT target neural network can calibrate to the DD material. Following this, the same process will be applied to calibrate towards the CD target material, which is typically used in the DCI experiments [1-2]. Although nuclear reactions of these materials have the same theoretical equation of state, the reaction cross-section involves the deposition of different types of particles. Our results show that after transfer learning, the original neural network can effectively predict the pellet productivity of various materials within an accepted error range. However, it should be noted that the increase of surface density and temperature brings about more nonlinear processes in the flow within the target pellet. In our future work, we will leverage experimental data from upcoming Double-cone Ignition experiments to further optimize our deep neural network, consequently enhancing our ability to accurately predict outcomes of future experiments.