Research On Chinese Herbal Medicine Plant Image Retrieval Based On Deep Learning And Hashing Learning

Due to the different growth periods,growth backgrounds,and shooting angles of Chinese herbal plants under natural scenes,traditional Chinese herbal medicine plant image retrieval systems are not very effective in retrieval.In order to solve the problem of low precision in the traditional Chinese herbal plant image retrieval algorithm,this paper applies deep learning technology and hash technology to the retrieval system.In terms of image semantic feature learning,the deep neural network has a strong ability to extract effective features from training data.Using its advantages in the extraction of image semantic features,it learns features similar to human's image visual semantic features,and these better image semantic features will help improve the recognition rate and retrieval rate of Chinese herbal medicine plant images.In the search mechanism,a hash algorithm was introduced to speed up the retrieval.The use of these two techniques not only ensures the extraction of human visual semantic features,but also improves the real-time and retrieval accuracy of the search.The major contributions of this work are:1.The related theories and key technologies of traditional image retrieval algorithms are studied in depth in this paper,and the advanced algorithms are introduced in detail.The problems that the current image retrieval technology of Chinese herbal medicine faced are pointed out,including the high dimension of extracted features,and the "semantic gap" between the features extracted by machine and the image content understood by human.2.The basic principles of deep learning including forward propagation,backward propagation,convolutional layer and pooled layer,activation function,loss function,etc.are introduced in detail.And analyze the deep neural network from the theoretical perspective and verified its strong ability to learn image features.3.For "dimensional disaster" and "semantic gap",this paper proposes a supervised model based on deep learning and hash learning.In feature extraction of Chinese herbal medicine images,we used the deep learning network to efficiently learn the feature expression of Chinese herbal medicine images.At the same time,a hash layer is added behind the deep learning network to directly generate a binary hash code and reduce the quantization error.4.Designed and developed a Chinese herbal plant image retrieval system based on deep learning and hash learning,including the establishment of image high-dimensional features library,establishment of binary hash code base,query upload,and retrieval of similar images.