Classification of Flowers under Complex Background Using Inception-V3 Network

In recent years, benefiting from the introduction of deep learning network algorithms such as Inception, Resnet, and Mobilenet, the accuracy of object classification has been significantly improved, especially for flower classification. Furthermore, with the development of mobile terminals, it becomes common for non-professional people to take photos of wild flowers, which makes flower classification an attractive feature. However, due to the blur effect of photos, it is challenging to achieve high accuracy and robustness in terms of classification. In this paper, we propose a three-step automatic classification scheme based on Inception network. We first preprocess the flower image to filter out blurred images. Then, the images in the training set are segmented by GrabCut, and the flowers are segmented by background to increase the number of samples in the training set. Then, we adopt the Inception-V3 network to extract the features of clear images and perform classification. The results show that the proposed scheme can improve the classification accuracy rate by a maximum of 40.35 %, reaching 97.78 %.