Adaptive Sparsification of Edge Vision Transformers for Artificial Intelligence Agriculture Pest Scouting Under Thermal Drift: A Computer Accuracy-Latency Regret Evaluation

Edge vision transformers for pest scouting in artificial intelligence agriculture must operate under thermal drift, where device temperature alters latency and accuracy in ways that static sparsification cannot anticipate. the present analysis formalises adaptive sparsification as a regret minimisation problem over a sequence of thermal states and proposes a threshold-based controller that adjusts token pruning rates in response to observed latency and confidence signals. Because no field data are available, the operating characteristics of the evaluation design are characterised through a replication-based power analysis 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 for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The analysis indicates that 52 clusters are required to detect a 5-percentage-point effect at 80% power. The contribution is a formal definition of accuracy-latency regret for adaptive sparsification and a demonstration that the before-after-control-impact design can identify the controller's effect under realistic clustering. The framework provides a basis for evaluating whether adaptive sparsification reduces regret relative to fixed pruning under thermal drift, but it does not report empirical performance.