Feature Extraction Algorithm And Application Based On Deep Hashing

At present,with the development of image acquisition technology,image-related applications are becoming more and more widespread.Since the amount of information contained in the image is large,extracting distinguishing abstract features from the image is a key step in object recognition.Traditional image features are hand-designed features that rely on the designer's prior knowledge and tend to be narrower.In the era of artificial intelligence,deep learning is widely used to learn the feature representation of images.Deep learning based methods can learn feature representation from a large amount of data,and obtain features with strong expressiveness and generalization ability.Based on the above research background,this paper studies the feature extraction algorithm based on deep hashing.The deep hashing-based algorithm is one of the best retrieval algorithms at present.This algorithm uses the related techniques in the depth learning field to learn the parameters of the hash function,and maps the original high-dimensional data of the image into a series of binary codes,which simplifies the feature.The main work of this paper is as follows:1.This paper studies and summarizes the key technologies of deep hashing network,and studies the image feature extraction method based on HashNet network.This method can extract features from the original image and generate a compact hash code for feature expression,which solves the problem of gradient disappearance and data imbalance in deep hashing network training.2.The bilinear pooling technique for fine-grained classification of images is studied.Bilinear HashNet is proposed to generate binary features of images.This network replaces the fully connected layer with a bilinear pooling module,replaces the feature encoding method,and realizes fine-grained retrieval and classification of images by using the relative positional relationship between features.The experimental results show that the Bilinear HashNet has a significant performance improvement over the HashNet network.3.The vehicle recognition system based on Bilinear HashNet is implemented on the Caffe framework.A phased training method and a data augmentation method are proposed to improve the network model for ease of use and performance.