An efficient fruit fly neural network Hash search method for WMSN data

Abstract

The invention discloses an efficient fruit fly neural network Hash search method for WMSN data, and the method comprises the steps: firstly, carrying out the preprocessing of a data set, enabling thecharacteristics of the data set to be converted into numerical data, and enabling the Fast Johnson-to be used for carrying out the Hash search of the WMSN data; the Lindenstrauss Transform (FJLT) projection matrix projects data to a measurement space with a higher dimension, so that the content similarity between original data objects is ensured, and accurate search is facilitated; and finally, awin-win characteristic selection strategy in a fruit fly olfactory neural local sensitive hashing method is adopted to reduce the dimension of the data set and improve the search efficiency. When thefruit fly olfactory nerves are used for simulating the locality sensitive hashing process, better universality is achieved, the accuracy of the search result is improved, the approximate neighbor query problem of high-dimensional big data is effectively solved, and the method is effectively applied to WMSN data search affairs based on a WMSN application system.

Description

technical field [0001] The invention relates to a wireless multimedia sensor, in particular to an efficient fruit fly neural network hash search method for WMSN data. Background technique [0002] Wireless Multimedia Sensor Networks (WMSN) is a new type of wireless sensor network with video, audio, image and other multimedia information developed on the wireless sensor network (WSN). So far, WMSN has been widely used in security monitoring, intelligent transportation, environmental monitoring, etc. WMSN multimedia data query is an important core technology in the research and development of WMSN application systems. WMSN multimedia data usually has the characteristics of high dimensionality, large scale, and multiple types. Conventional neighbor search algorithms cannot meet the needs of the system. hot issues of concern. [0003] Usually, people need to perform dimensionality reduction processing on WMSN multimedia data. Generally, there are two types of dimensionality re...

AI Technical Summary

Problems solved by technology

[0006] Although the FLSH method proposed by Sanjoy and other scholars in the journal Science in 2017 has many advantages in large-scale high-dimensional data search, there are certain problems in the local sensitive hash search process based on random projection contained in the middle package of the FLSH method.

[0007] Question 1 (the trade-off between search result accuracy and time efficiency): In the process of high-dimensional data neighbor search, we often need to weigh the accuracy of search results and the accuracy of search time, although the local sensitivity based on random projection The Greek strategy reduces the time complexity of neighbor search, but at the same time loses some accuracy;

[0008] Question 2 (hidden connection between data set processing and neural perception): The traditional approximate nearest neighbor search based on locality-sensitive hashing is to process data objects on the search target (data set), and the subject who initiates the query action identifies the data object The activity of sensory neurons when the time is often overlooked by researchers

[0009] Question 3 (Universality of the J-L theorem in dimensionality reduction of high-dimensional data sets): The J-L theorem has been widely cited in the field of data dimension reduction, but about the new features of the data set after the FJLT matrix performs spatial transformation on the data set The distribution law of the new features on the data points is still to be studied

We found that the sparse binary random matrix it uses is not suitable for low-distortion data projection [7], although the speed of data projection is improved, but at the same time, the partial similarity between data objects is sacrificed as a cost, so it cannot meet the WMSN application system. Requirements for query accuracy; we also realized in the research that the stability of the FLSH algorithm needs to be further improved, and the query accuracy of the algorithm on different data sets is quite different, which is not suitable for the query mechanism in the WMSN application system; finally, the The feasibility of the algorithm lacks strong theoretical support. Although the sparse binary matrix can better simulate the activation mode of neurons, it lacks corresponding theoretical support in the process of data set processing.

Images