AI and Non-Technical Survey: Quantifying the Impact of AI-augmented Drone Imagery Analysis

This paper examines the impact of AI‑augmented drone imagery analysis on the efficiency and accuracy of Non‑Technical Survey (NTS) operations conducted by Norwegian People’s Aid (NPA) in Ukraine. The study focuses on the integration of AI‑augmented imagery analysis, using SpotlightAI, into drone‑based NTS workflows, comparing traditional manual imagery analysis with AI‑assisted approaches. Using a mixed‑methods design that combines longitudinal trend analysis (November 2023–February 2025) and a six‑week quasi‑experimental study in 2025, the research evaluates potential effects on person‑hour efficiency of confirmed hazardous area (CHA) identification and evidence detection accuracy.

The findings indicate that although AI‑augmented imagery analysis does not significantly increase the absolute speed of completing NTS tasks, use of AI-augmented imagery analysis corresponds positively with greater efficiency and accuracy. In the quasi‑experimental study, the team using AI-augmented imagery analysis identified more than double the CHA area per person‑hour compared to the team which did not use AI.

Furthermore, using the percentage in size of CHAs identified within the total area of assigned sites/ villages tasked for investigation by NTS teams as a proxy for accuracy of direct evidence identification, the findings indicated that the use of AIaugmented imagery analysis produced up to 2.85 times greater NTS accuracy. Additionally, it was noted that over a one‑year operational period, NTS conducted with the use of AI-augmented drone imagery analysis accounted for 76.4% of all CHA identified by NPA in Ukraine despite being used on only 13.4% of sites, suggesting a disproportionate contribution to overall land release operations.

While the paper does not establish causal mechanisms or statistically rigorous correlations, it provides empirical evidence supporting the potential value of AI integration in mine action and underscores the need for further research with larger samples and more granular activity‑based time analysis.