Flower Image Classification Model And Algorithm Based On Transfer Learning

In the field of botany,the classification of flower families and genera is a very basic work.Different types of flowers may have similar characteristics such as for example color,shape and appearance.For ordinary people who do not have a deep understanding of plant and flower-related knowledge,it is still difficult to accurately distinguish common types of flowers.In this regard,this article uses deep learning based convolutional neural network algorithm to realize the recognition and classification of a variety of common flowers.The main research contents of this article are:Collect and sort out a flower data set containing 20 kinds of flower images.Part of the data set comes from the Oxford 102 Flowers data set,and the other part is obtained by the web crawler.The image data set is augmented by flipping,rotating and adding noise.The augmented data set is divided into training set,validation set and test set according to a certain proportion.Use self-built convolutional neural network to realize flower image recognition.An improved optimization algorithm that combines Adam with the learning rate decay strategy is proposed.After experiments,the accuracy of the improved optimization algorithm is 2%higher than that of the model trained with the Adam optimization algorithm.The migration learning based on Alex Net and VGG16 pre-training model realizes flower image classification.Two different migration learning strategies were used to perform migration learning training on Alex Net and VGG16,and the loss and accuracy of the model on the training set and the validation set during the training process were compared.The experimental results show that the transfer learning method based on the Alex Net pre-training model network model using feature extraction and fine-tuning the output layer can achieve90% accuracy in the test set,while the VGG16 network model uses the three-layer fully connected layer transfer learning method after fine-tuning on the test set It can reach 92%accuracy.A flower image recognition and classification method based on migration learning model fusion is proposed.Use GAP to replace the traditional fully connected layer,improve the Alex Net and VGG16 network models,and perform migration training on the flower image data.Compared with the migration training method based on the original pre-training model on the flower image data,the improved model has improved model accuracy and training efficiency.In order to further strengthen the generalization of the model,the experiment combines the idea of model fusion in ensemble learning,and uses the simple average method to fuse the three models.It is verified by experiments that the fusion integrated model has effectively improved the recognition accuracy and stability of the flower image data set compared with a single model.