In magnetic confinement fusion experiments, ion temperature is one of the key parameters affecting the fulfillment of Lawson’s criterion, the rate of reaction generation and the quality of confinement. Currently, the commonly used methods to diagnose ion temperature in magnetic confinement fusion devices include Langmuir probe, charge exchange recombination spectroscopy (CXRS) method, laser-induced fluorescence method, neutron yield and energy spectrum method, etc. However, these methods are generally characterized by the shortcomings of not being able to measure the ion temperature of the core, or needing to occupy the optical diagnostic window, or not being able to measure the temperature in real time, or only being able to measure the temperature of the plasma in the thermally-equilibrium state. How to diagnose the ion temperature of the core in future closed fusion reactors without optical diagnostic windows has become an important issue in Tokamak operation. By utilizing the characteristics of neutral beams mainly deposited in the plasma core and strong penetration ability of fusion neutrons, we establish a method for diagnosing core ion temperature using fusion neutron yield coupling under neutral beam injection conditions. This method can self consistently distinguish the contributions of thermonuclear neutrons and beam target neutrons, thereby achieving real-time measurement of ion temperature. First, the evaluated nuclear data file (ENDF) database is used as the baseline to perform fine velocity-group calculations. A temperature-dependent correction is then applied to the D–D fusion reaction cross section. This process yields a multi-temperature differential cross section database. Then, the ionization and slowing down process of deuterium atoms is calculated, which leads to the formation of a database of the neutron yields for the thermonuclear and the beam-target reaction. Finally, an iterative algorithm is designed to calculate the respective yields of the two types of neutrons and the ion temperatures in real time. The effects of yield and density input parameters on the applicability of the method were evaluated. Their influence on uncertainty amplification was also assessed. This method provides data support for the operation and parameter optimization of the Tokamak device. The method is demonstrated and validated using experimental data from the HL-3 tokamak.