CERASUS HUMILIS CULTIVARS IDENTIFICATION WITH SMALL-SAMPLE AND UNBALANCED DATASET BASED ON EFFICIENT NET-B0+RANGER NETWORKS

Because of the high similarity leaves different Cerasus humilis varieties, it is difficult to identify them with naked eye. In this study, four varieties collected in field were used as research objects, and a new leaf recognition model based on improved lightweight convolution neural network EfficientNet-B0 was proposed. Firstly, performance models Efficientnet-B0 ResNet50, GoogleNet, ShuffleNet, MobileNetV3 compared two learning methods. Then, influence optimizers accuracy optimal model. Finally, rates optimize The results show that rate proposed +Ranger+0.0005 up 86.9%, which 2.23% higher than original method can effectively improve auriculate leaves, provide reference for deployment identification variety mobile terminal.