The research niche of this article is the use of open-source visual intelligence and automated computer vision, combined with classical time-series modelling, to analyse and forecast military equipment losses in the Russo-Ukrainian war. The purpose of the research was to test whether visually confirmed data (the Oryx repository) can be algorithmically transformed into a reliable weekly series and used for short-term forecasting. Two hypotheses were examined: (H1) open visual data yield a series with stable trend and seasonality; (H2) ARIMAX with Fourier terms outperforms seasonal ARIMA at short horizons. The methodology comprised a Python pipeline in which YOLOv8 localized date stamps and EasyOCR read them; STL decomposition characterized the trend and seasonal structure. Forecasting employed SARIMA and ARIMAX with sine/cosine pairs for the annual period. Results confirm pronounced annual seasonality (peaking in March–April) and a trend cresting in early spring 2023; relative to SARIMA, ARIMAX reduced errors by 14.6–23.7% (MAE) and 23.8–25.5% (RMSE) in both in-sample fit and rolling validation. The conclusions indicate that, despite limitations (a lower bound on true losses), public visual data provide a robust, verifiable analytical basis with strong predictive potential; future work should incorporate machine learning, exogenous covariates, and probabilistic forecasting.