Air pollutants of PM2.5 are fine particles having a significant impact on human health. Recently, Taiwan Environmental Protection Agency has been promoting the installation of air pollution quality micro-sensors to monitor PM2.5 concentrations at various locations in Taiwan. However, the measurements from PM2.5 micro-sensors are easily influenced by instrument anomalies or the nearby environment, and the number of micro-sensors is too large to be investigated individually. In this study, we propose a two-stage approach to analyze such data. In the first step, a spatio-temporal model was established to describe both the spatial and the temporal signals, and to reduce the noises in the raw observations. The second step then grouped the micro-sensors into clusters based on a functional principal component analysis (FPCA). The main advantage of this two-step method is its computational efficiency. Moreover, the resulting functional principal components take into account the spatial correlations between micro-sensors, which were not considered in a typical FPCA. We investigated the time series of PM2.5 concentrations measured by the micro-sensors, and we identified several regional clusters representing different characteristics of residents' activities or industrial land uses.