Reliability-calibrated contrastive learning improves plant leaf recognition
Prototype-reliability calibration improved plant leaf recognition across three benchmarks and a laboratory-to-field test while leaving the exported classifier architecture unchanged.