Adversarial Patch Robustness Certification for Artificial Intelligence Agriculture Pest Detection Under Field Occlusion: A Computer Certified-Radius Evaluation

Adversarial patches placed in a crop canopy can suppress or redirect the detections of a convolutional pest detector, and the field occlusion that motivates such patches also degrades clean-image accuracy, so a robustness claim that ignores occlusion is not a field claim. No observational data are used or claimed: the analysis characterises the operating characteristics of the design under a declared variance structure with between-cluster SD = 6.0, residual SD = 4.5 and intraclass correlation = 0.640, across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effects of 3, 5 and 8 percentage points. The framework separates the certified radius of a detection rule from the occlusion-induced radius contraction, and it shows that the two compose additively in the certified bound. The design study indicates that 52 clusters are required for 80% power at a 5 percentage point effect under the declared variance structure, and that the minimum detectable effect at 80% power falls as cluster count rises from 8 to 32. The contribution is a certification object that a field evaluation can be powered against, rather than a claim about any particular deployed detector.