An Online Hash Nearest Neighbor Query Method Based on Data Block Learning

Abstract

The invention discloses an online hash nearest neighbor query method based on data block learning, which is characterized in that it includes the following steps: acquiring and preprocessing image data, defining a hash model for processing data, establishing and judging whether the updated hash vector is reasonable The Hamming distance predicts the loss function, obtains the objective function, optimizes the objective function, and performs online hash nearest neighbor query on the given data to be queried in the test database; the advantage is that it is mainly based on the idea of ​​​​data blocks, and each time processing stream small data block, and design an optimization algorithm in a smaller data space to improve learning efficiency. In the design method, the Hamming space loss between data samples inside the data block is guaranteed to be minimal, and at the same time, the incremental change of the overall online learning is also constrained. , the noisy data is effectively controlled, the query efficiency of the given data to be queried is greatly improved, and the overall storage space is reduced.

Description

technical field [0001] The invention relates to an online nearest neighbor query method, in particular to an online hash nearest neighbor query method based on data block learning. Background technique [0002] Nearest Neighbor Search is an important research direction in the field of information retrieval, and it is widely used in image retrieval and data mining. In the hash-based nearest neighbor algorithm, the original data is mapped to the binary code in the Hamming space through the hash function, and the Hamming distance coding is used to be as close as possible to the original input space data, and the XOR operation using the speed of the computer close to the hardware has high efficiency. It has the advantages of data processing and search efficiency, so it is widely used. In practical applications, there are various forms of data, most of which are dynamically generated, such as the amount of new web pages, flight and railway traffic, and weather condition informat...

AI Technical Summary

Problems solved by technology

Therefore, for large-scale data, because the hard disk space is much larger than the memory, it is impossible to read all the data into the memory, and it is also very time-consuming to process and calculate a large amount of data at one time.

However, most of the current methods are based on the hash learning method of batch processing technology, which is equivalent to considering all the data and retraining the hash function when new data arrives, which is inefficient for streaming dynamic data

[0003] At present, online hash learning nearest neighbor search based on a pair of data, although the update frequency and stability of the learned hash has been improved, but the actual processing mechanism is still updated based on a pair of data points, in order to speed up the hash function Update efficiency, propose a nearest neighbor query method based on online hash learning on block data

[0004] The goal of online hash learning is to process streaming data sequentially, but it relies on current training data. There have been a lot of related algorithm research, but there is less research on online hash functions based on data block learning.

Images