In severe accidents of nuclear power plants (NPPs), particle of fission products (FPs) included in air and vapor would be released from the reactor vessel, through the water to the atmosphere. In these situations, most of the FPs are considered to be trapped inside the water, which means that the air is been decontaminated. This effect is called pool scrubbing. Although pool scrubbing is important in evaluating the effect of NPP accidents, its mechanism has not been explained due to its complexity in flow structure. In codes such as MELCOR, used in severe accident analysis, the primary factor of decontamination by the pool scrubbing is defined as the inner flow of particles inside each single bubble of the two-phase flow. On the other hand, there is no available experimental evidence of this phenomenon. Therefore, in this paper, interface velocity of an oil droplet, as a simulated fluid of air bubble, is measured experimentally by using the combination of PIV and visualization. To obtain the decontamination of aerosol from the bubble, the Mach-Zehnder interferometer is used to measure the density field of aerosol decontamination from the bubble interface. To clearly capture density profile in the vicinity of bubble, Machine-learning-based data processing is applied to improve the ability of phase data extraction. With the density field compared to the data of interface velocity, obvious relationship of phase field and aerosol decontamination behavior is confirmed.