A Non-Destructive Method for Identification of Tea Plant Cultivars Based on Deep Learning

Tea plant cultivar identification is normally achieved manually or by spectroscopic, chromatographic, and other methods that are time-consuming often inaccurate. In this paper, a method for the of three tea cultivars with similar leaf morphology proposed using transfer learning five pre-trained models: EfficientNet-B0, MobileNetV2, MobileNetV3, MobileViT-S, ShuffleNetV2. The results showed best test accuracy percentages ShuffleNetV2 were 98.33, 99.67, 99.33, 98.67, 99.00%, respectively. most lightweight model was ShuffleNetV2, fastest combination 112 × image resolution. Considering accuracy, number parameters, floating point operations (FLOPs), MobileNetV2 not only accurate model, but also both fast. present research could benefit farmers consumers via identifying without destructive techniques, factor would reduce adulteration commodity tea.