Benchmarking Evaluation of Bayesian Deep Homography Estimation for Computer Agriculture Aerial Weed Mapping Under Rolling-Shutter Distortion: An Artificial Intelligence Georeferencing-Error Bound for Spray-Drift Liability

Aerial weed mapping for site-specific spraying depends on homography estimation that converts oblique camera imagery into georeferenced field coordinates, and rolling-shutter distortion introduces a systematic, speed-dependent bias that propagates into spray-drift liability. The framework combines a formal error decomposition, a Bayesian homography estimator with calibrated uncertainty, and a liability bound expressed as a function of the posterior variance of the homography parameters. A replication-based evaluation of the before-after-control-impact estimator was conducted with no observational data: the analysis characterised the operating characteristics of the design under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64). Power was estimated across 2000 design cells. The design parameters were clusters of 8, 16, 24 and 32, and effect sizes of 3, 5 and 8 percentage points, with 2000 replications per cell. The computed intraclass correlation was 0.64. The framework proposes that the georeferencing-error bound scales with the square root of the posterior variance of the homography parameters and with the rolling-shutter line-readout time, and that the bound can be used to set a conservative spray-buffer width. The contribution is a benchmark protocol and a formal liability bound that together allow Bayesian deep homography estimators to be compared on a quantity that matters for spray-drift regulation rather than on reprojection error alone.