Research And Implementation Of Flower Image Retrieval System Based On Deep Learning

With the continuous progress of science and technology,the number of digital images grows at an amazing speed.In the face of a huge digital image database,how to quickly and effectively find the image users need has become a hot topic of research.The traditional method of text-based image retrieval has been unable to meet the requirements of users.Nowadays,content-based image retrieval has developed rapidly.In order to better understand the resources of flowers,the retrieval of flower images is a job with good economic value and social value.This paper introduces the present situation and core technology of content-based image retrieval at home and abroad.In recent years,deep learning has made remarkable achievements in computer vision research.Based on deep learning,this paper used the convolutional neural network model to extract the characteristics of the flower image and studied the flower image retrieval.The main work is as follows:(1)In order to extract flower image features with better effect,this paper compared different convolutional neural network models and selected the model based on vgg-16 to extract image feature vectors.In this paper,an optimization method is proposed for vgg-16,in which two feature pyramid multi-layer fusion modules and residual attention mechanism modules are introduced.In the deep layer of the original network,there is abundant high-level semantic information but lack of spatial location information.In the shallow layer,the opposite is true.Therefore,the introduction of multi-layer fusion module of feature pyramid can effectively integrate high-level and low-level features,so that the two information complements each other to get better image features.In the process of image feature extraction,the whole flower image is used to extract the global features,and the weeds,soil and stones in the background of the flower image must be included,which will have a lot of redundant information and affect the accuracy of retrieval.The introduction of the residual attention mechanism module can effectively eliminate redundant information and optimize the feature graph.Adding these two modules into vgg-16 can effectively optimize the extracted image features.(2)Different index methods are compared,such as index methods based on tree structure,high-dimensional data clustering and hash algorithm.The existing hash algorithm with good effect is selected.In this paper,the feature mapping of the second layer in the full connection layer is selected as the feature vector of the image.The retrieval algorithm combining principal component analysis and hash search is selected.Three classical hashing algorithms,locally sensitive hashing,translation-invariant kernel locally sensitive hashing and density-sensitive hashing,were selected for hashing retrieval.When using PC A to reduce the dimension of feature vector to 64 and the encoding length to 256 bit for locally sensitive hashing,the retrieval algorithm had the best effect.The validity of the combination of PCA and hash retrieval is verified.(3)In this paper,the improved VGG-16 extraction feature,PCA and local sensitive hash combination indexing method are used as the model and algorithm of flower image retrieval.The Oxford 102 Flowers data set is used for verification.The experimental results show the recall rate and investigation of the algorithm.The quasi-rate is high.The superiority of the improved VGG-16 model is demonstrated by comparison with other models and methods of feature extraction and the original VGG-16 model.Finally,a practical flower image retrieval system is realized by the method of this paper.