Event-Triggered Edge Inference for Artificial Intelligence Agriculture Pest Monitoring Under Intermittent Connectivity: A Computer Communication-Energy Regret Bound

Edge inference for automated pest monitoring in agricultural fields is constrained by intermittent wireless connectivity and finite battery budgets, yet the design of triggering policies that trade communication against detection quality remains largely heuristic. the present analysis formalises event-triggered edge inference as an online decision problem in which a sensor node chooses, at each time step, whether to transmit a local inference to a gateway or to act on it locally, under a stochastic communication cost and a detection-loss penalty. A regret bound is derived for a threshold-based triggering policy relative to an oracle that knows the connectivity process in advance. The bound decomposes into a communication-energy term that scales with the square root of the horizon and a detection-loss term that scales with the inverse of the triggering threshold, and it holds under a bounded-drift assumption on the connectivity process. The analysis is illustrated with a design-characterisation study: a before-after-control-impact estimator is evaluated across 8, 16, 24 and 32 clusters, effect sizes of 3, 5 and 8 percentage points, and 2000 replications per cell, under a declared variance structure with between-cluster SD of 6.0, residual SD of 4.5 and intraclass correlation of 0.64. The design-characterisation results indicate that 54 clusters are required to detect a 5-percentage-point effect at 80% power, and that the intraclass correlation of 0.64 dominates the power profile. The regret bound provides a principled basis for setting triggering thresholds, and that the design-characterisation results bound what any field evaluation of such a policy can credibly detect.