A Flower Classification Framework Based on Ensemble of CNNs

Currently, the classification of flower species has become a hot topic in the field of image classification. Flower classification belongs to the category of fine image classification, and such images are usually represented by multiple visual features. At present, all the flower classification methods based on a single convolutional neural network (CNN) model can hardly extract the features of a flower image as much as possible. In view of the limitation of description methods for flower features and the problem of low accuracy of flower species recognition, this paper proposes a flower classification framework based on ensemble of CNNs. The method consists of the following three parts: (1) The same flower image is processed differently to make the color, texture and gradient of the flower image more prominent; (2) Fine-tune the structure and parameters of the convolutional neural network to adapt it to the extraction of corresponding features. Then use the CNN model with different characteristics to extract the corresponding features; and (3) A framework that can fuse each CNN sub-learner is used to combine various features effectively. We tested the effectiveness of our method on the Oxford Flowers 102 Dataset [2]. The result demonstrates that the proposed approach effectively improves the accuracy of flower classification.