Benchmarking Evaluation of Spiking Neural Networks for Computer Agriculture Edge Pest-Detection Under Thermal-Sensor Power Constraints: An Artificial Intelligence Energy-Accuracy Trade-off Audit for Field-Deployment Safety

Edge pest detection in computer agriculture is constrained less by classifier accuracy than by the thermal power budget of the sensing node. Spiking neural networks (SNNs) promise order-of-magnitude energy reductions on neuromorphic hardware, but the field lacks a design-level account of how that promise translates into deployable detection performance when the thermal sensor itself is the dominant load. The framework formalises a thermal-power budget, defines an energy-normalised detection rate, and derives a power-accuracy frontier for leaky integrate-and-fire classifiers operating on low-resolution thermal frames. A replication-based evaluation of the before-after-control-impact estimator was conducted under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64), with power estimated across 2000 design cells. The analysis indicates that 54 clusters are required to detect a 5-percentage-point effect at 80% power. The contribution is a reproducible audit protocol that makes the energy-accuracy trade-off explicit and falsifiable before field deployment, and that identifies the sensor duty cycle, not the network, as the binding constraint in most edge configurations.