The ESA PLATO space mission is devoted to unveiling and characterizing new extrasolar planets and their host stars. This mission will encompass a very large field of view, granting it the potential to survey up to one million stars depending on the final observation strategy. The telemetry budget of the spacecraft cannot handle transmitting individual images for such a huge stellar sample at the right cadence, so the development of an appropriate strategy to perform on-board data reduction is mandatory. We employ aperture photometry to produce stellar light curves in flight. Our aim is thus to find the mask model that optimizes the scientific performance of the reduced data. We considered three distinct aperture models: binary mask, weighted Gaussian mask, and weighted gradient mask giving lowest noise-to-signal ratio, computed through a novel direct method. An innovative criterion was adopted for choosing between different mask models. We designated as optimal the model providing the best compromise between sensitivity to detect true and false planet transits. We determined the optimal model based on simulated noise-to-signal ratio and frequency of threshold crossing events. Our results show that, although the binary mask statistically presents a few percent higher noise-to-signal ratio compared to weighted masks, both strategies have very similar efficiency in detecting legitimate planet transits. When it comes to avoiding spurious signals from contaminant stars however the binary mask statistically collects considerably less contaminant flux than weighted masks, thereby allowing the former to deliver up to $\sim$30\% less false transit signatures at $7.1\sigma$. Our proposed approach for choosing apertures has been proven to be decisive for the determination of a mask model capable to provide near maximum planet yield and substantially reduced occurrence of false positives.