A Massive Image Classification System Based on Deep Hierarchical Feature Learning

Abstract

The present invention provides a massive image classification system based on deep hierarchical feature learning, comprising the following steps: step a, inputting unlabeled and labeled image data, preprocessing the unlabeled image data, removing interference information, and retaining key information; Step b, perform K-means feature learning on the preprocessed image to obtain the dictionary of this layer; step c, if this layer is the Nth layer, perform feature mapping on the dictionary of this layer and the labeled image data set to obtain a deep layer After the feature, proceed to step e, otherwise perform feature mapping on the dictionary of this layer and the unlabeled image data to obtain deep-level features; step d, according to the correlation of deep-level features, aggregate multiple high-correlation features into a receptive field , if this layer is N-1 layer, then proceed to step e, otherwise it will be sent to step b as the input information of the next layer; step e, in the N layer, input the learned features into the SVM classifier for classification.

Description

[0001] Invention field [0002] The invention belongs to the technical field of machine learning and image processing, and relates to massive image processing on a distributed platform, in particular to an implementation scheme for massive image classification based on depth-level features. Background technique [0003] Nowadays, with the development of multimedia technology, a large number of multimedia data including images, audio, video and other information have emerged. How to classify a large amount of information has become a hot issue in multimedia technology research. Image classification research tasks are mainly pre-processing, Feature extraction and classification are composed of three main links, and each link has an important impact on the classification effect of images. With the rapid development of computer software and hardware and Internet technology, the amount of multimedia data has also increased at an alarming rate. More and more information in the industry i...

AI Technical Summary

Problems solved by technology

[0003] Nowadays, with the development of multimedia technology, a large amount of multimedia data including images, audio, video and other information emerges. How to classify a large amount of information has become a hot issue in the research of multimedia technology. The research task of image classification is mainly composed of preprocessing, Feature extraction and classification are composed of three main links, and each link has an important impact on the classification effect of images. With the rapid development of computer hardware and software and Internet technology, the amount of multimedia data is also increasing at an alarming rate. In the industry, more and more information is expressed in the form of images, which undoubtedly brings great challenges to all aspects of the task of image classification.

Traditional image classification is carried out on a single machine by extracting color, texture, and shape features. With the continuous increase of the image library, the complexity continues to increase, and the artificially designed features of the single machine are far from meeting the demand. Using parallel Dealing with is undoubtedly a good solution

Images