Modeling the spatial dynamics of weed communities based on unmanned aerial vehicle (UAV) imagery can contribute to evaluating colonization process with temporally and spatially high resolutions. The hidden Markov model is likely to be suitable for describing the transition of spatial occupancy because it can be modified flexibly according to the object and can include the probability of classification error. This study aimed to infer the parameters of kudzu’s spatial dynamics and to evaluate its expansion after mowing through aerial image classification and site occupancy models. We obtained UAV images of the riverbank community and identified the grids (0.25 m²) occupied by kudzu with supervised classification. The hierarchical model comprised a model representing the uncertainty of the classification and a model representing the transition of the occupancy state of each grid. The posterior distribution of the parameters inferred with Markov chain Monte Carlo (MCMC) methods indicated that the expansion rate after mowing in July was higher than that after mowing in August. This suggests that compensation for mowing damage is relatively slow after August. The combination of image classification and hierarchical modeling would enable us to simulate weed spatial dynamics on a management scale and optimize management to reduce the impact of widespread invasive weeds.