Research On Fine-Grained Classification Of Flower Images

With the rapid development of computer technology,people can shoot various flower images through terminal devices easily and quickly such as mobile phones or cameras.However,the classification and recognition of flower images are not available to everyone and usually require professional knowledge and with the guidance of experienced experts.Fine-grained classification of flower images is one of the important research directions in the field of image classification.But,due to the complexity of flower growing environment and the diversity of flower itself,the traditional method and algorithm model of common object classification can’t solve the problem of the flower fine-grained classification.In view of the complicated background information of flower images and the diversity of flowers themselves,this paper studies the flower fine-grained classification from the two aspects of the salient region detection of flower images and the discriminant feature is extracted by using the double stream convolutional neural network.The main research work is as follows:In view of the complex background information of flower images,this paper presents a method of fine-grained classification of flower images based on salient regions.The method calculates the saliency map of flower image through the visual attention model,calculates the discriminant salient region of the flower image according to the saliency map of the flower images,extracts the features according to the salient region,and finally realizes the flower image automatic classification with the SVM classifier.Compared with the current method of manual marking the area of flower discriminant and the method of segmenting the flower images by segmentation algorithm,this paper proposes that the method of flower fine-grained image classification based on salient regions not only eliminates the manual identification of discriminative regions but also avoids the process of flower images segmentation,enhances the adaptability of the algorithm to the task of fine-grained classification of the flower images with complex backgrounds.In view of the diversity of the flower images themselves,very similar in intra-class and the big differences in inter-class.In this paper,we propose a double stream convolutional neural network model based on salient region.The traditional convolutional neural network lacks further comparative analysis of discriminative regions of flower images.However,the feature extraction based on salient regions for fine-grained classification of flower images lacks the analysis of flower background information.Therefore,this paper presents a double convolutional neural network model based on salient region,a double stream structure design,which runs independently of each other,and uses the convolutional neural network for the original image and the salient areas of flower image.Fusing the features extracted from the two layers of convolutional neural networks for fine-grained classification of flower images.The method can overcome the problem of small differences in inter-class and improve the accuracy of flower images with large differences in intra-class.The experimental results show that the proposed method of fine-grained classification based on salient region is more effective than the manual marking the discriminative regions of flowers and the classification of images using segmentation algorithm.The double stream convolutional neural network model based on salient region is more effective than using of the end-to-end convolutional neural network for the classification of flower images directly and extracting features based on salient region of flower image,can solve the task of fine-grained classification of flower images more effectively.