Research On Flower Classification Via Convolutional Neural Network

With the rapid development of electronic technology,it is convenient for people to take flower images by portable digital devices such as mobile phones,cameras,etc.And an effective way to automatically identify flower category has attracted more and more attention recently.Due to complex background,inter-class similarity and intraclass variation,it is hard to address the problems of natural flower classification by the traditional methods using handcrafted features.In this paper,a novel method of flower classification based on deep convolutional neural networks is proposed.We first preprocesses original image to determine the possible location of the flower.Then we utilize deep convolutional neural networks to automatically complete features extraction,and acquire more comprehensive characterization of flower images.Finally,a support vector machine(SVM)classifier is built with good features to accomplish flower classification.The main work of the paper is as follows:(1)Propose a new method of flower classification based on deep convolutional neural networks.Unlike traditional methods using hand-crafted visual features,our method utilizes convolutional neural network to automatically learn good features for flower classification and acquire more comprehensive characterization of flower images.(2)Propose a flower region slection method that preprocesses original flower image by combining its saliency map and luminance map.In general,flower images are taken in complex backgound.In order to get much better performance of flower classification,we adopt a flower region slection method that preprocesses original flower image by combining its saliency map and luminance map.The method can comprehensively utilize low-level features(such as color,brightness,etc.)and visual saliency in flower images,which achieves higher classification accuracy than traditional convolutional neural network.(3)Utilize support vector machine(SVM)classifier to accomplish flower classification.Due to the limitation of flower images in the paper,the training samples supplied to convolutional neural network are not sufficient enough to cause over-fitting.So a support vector machine classifier is built with good features to accomplish flower classification instead of Soft-Max layer in convolutional neural network.The Experimental results on our challenging flower dataset and Oxford 102 Flowers dataset demonstrate that our proposed method achieves higher classification accuracy than the previous known methods.