Abstract: The cloud field calibration of icing wind tunnels usually has the disadvantage of high instrument dependence. To solve this problem, this paper proposes a method for identifying the parameters of cloud fields in icing wind tunnels based on multi-modal fusion. This method takes the icing image of the test model together with the parameters such as the angle of attack, air velocity, air temperature, and icing duration of the model as input, extracts and fuses the two characteristic parameters, and takes the liquid water content （LWC） and the average volume diameter of water droplets （MVD） as the output to train the neural network model. And then the rapid identification of icing cloud parameters is realized. In order to verify the effectiveness and feasibility of the proposed method, the paper takes NACA0012 airfoil icing as the research object, develops the cloud field identification program of the icing wind tunnel, analyzes the influence of the fusion proportion, and obtains the best network model suitable for ice parameter identification. On this basis, simulation and experimental evaluation are carried out. The full scale error of the proposed method for LWC and MVD is less than 12%, which has high identification accuracy and good generalization performance, and can provide an important supplement for the identification of cloud fields in the icing wind tunnel.