Self-Supervised Acoustic Embedding Transfer for Artificial Intelligence Agriculture Grain Storage Insect Detection Under Low-Cost Microphone Degradation: A Computer Recall-Robustness Evaluation

Low-cost microphones deployed in grain stores degrade the acoustic signal that insect-detection models depend on, and the recall loss this causes is rarely quantified before deployment. the present analysis treats the problem as a design-characterisation question rather than a field trial: it specifies a self-supervised acoustic embedding transfer pipeline, defines a recall-robustness functional that maps microphone degradation to detection recall, and evaluates the operating characteristics of a before-after-control-impact (BACIP) estimator that a future field study would use to test the pipeline. The evaluation is a replication-based power analysis with no observational data. Under a declared variance structure (between-cluster SD = 6.0, residual SD = 4.5, intraclass correlation = 0.64), power was estimated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The analysis identifies 54 clusters as the number required for 80% power at a 5-percentage-point effect, and shows that the recall-robustness functional is estimable with acceptable precision only above that threshold. The contribution is a formal bridge between acoustic embedding transfer and the design of credible recall evaluations for smallholder grain-storage settings.