Flower Recognition Based on an Improved Convolutional Neural Network MobileNetV3

To improve the traditional flower image recognition method, poor accuracy, poor generalization function and long recognition process. In this paper, we mainly propose a flower classification and identification method based on the improved MobileNetV3. In this model, MobileNetV3, a lightweight convolutional neural network, was selected for pre-training on the Oxford 102 dataset, and then the shared parameters obtained from the pre-training were passed to the model for flower identification, and the model was further fine-tuned. Compared with Resnext152, AlexNet, the results showed that the improved MobileNetV3 had the best overall learning with recognition accuracy exceeding 99%.