Зонно-комбінаторний метод обробки зображень в системах технічного зору

Abstract

The cybersecurity of critical infrastructure is increasingly threatened by small unmanned aerial vehicles, necessitat-ing robust detection by machine vision systems. The primary challenge for machine vision systems is the high level of non-stationary background noise caused by atmospheric brightness dispersion, whose statistical parameters (mean and variance) fluctuate significantly depending on the time of day, often masking the weak signal from a dis-tant object. This paper presents a zonal-combinatorial approach for robust object detection, validated using a Py-thon-based simulation that generates 100x100-pixel frames corrupted by additive Gaussian noise with time-variable presets (e.g., 'day' vs. 'night') and embeds a 1x1-pixel target signal. The methodology consists of two stages: zonal processing, where various spatial filters are applied to suppress noise and enhance the target's Signal-to-Noise Ra-tio, and сombinatorial analysis, which makes a detection decision based on the filtered output. The main contribu-tion is a comparative analysis of different filtering algorithms and their efficacy in increasing the output SNR under dynamic noise conditions, which is essential for reliably isolating candidate "zones" (potential targets) for subse-quent spatio-temporal analysis and final threat classification in technical vision systems.