Feature extraction method for ocean multi-dimensional asymmetric signal with given observation dimension

Abstract

The invention discloses a feature extraction method for an ocean multi-dimensional asymmetric signal with a given observation dimension. The feature extraction method comprises the steps of ocean multi-dimensional asymmetric signal recombination, ocean multi-dimensional asymmetric signal segmentation, feature aggregation and dimension reduction considering sea temperature distribution structure differences, feature extraction based on reduction data and experimental verification. According to the method, effective information is extracted from mass data according to a given observation dimension while spatial-temporal heterogeneity and dimension asymmetry of the ocean signals are considered, and then feature extraction of the ocean multi-dimensional asymmetric signals is completed.

Description

technical field [0001] The invention belongs to the field of physical geography and marine science, and in particular relates to a method for extracting features of marine multidimensional asymmetric signals with a given observation dimension. Background technique [0002] At present, there are many kinds of marine data sampling methods, including marine remote sensing observation, seabed-based automatic observation, acoustic detection technology, etc. Satellites, ships, buoys and other carriers are used as observation platforms to realize the collection of marine multi-dimensional data. Affected by actual needs and sensor update iterations, the types of data collected by sensors continue to expand, and the volume of data continues to increase, forming massive multi-dimensional data. With the continuous deepening of data analysis requirements and the continuous subdivision of spatiotemporal granularity, stripping the characteristics of requirements from massive high-dimensio...

AI Technical Summary

Problems solved by technology

However, all dimensions of high-dimensional data are treated equally, which ignores the heterogeneity between different dimensions, which may lead to estimation bias; dimensionality reduction methods represented by data aggregation use the mean and median of data in a specific dimension Waiting to replace the data of the entire dimension to complete the reduction of the data dimension, the essence is to use local distribution characteristics to replace the data of the entire dimension

For data with large differences in distribution in space or time, this dimensionality reduction method often ignores the feature information in the reduced dimension, resulting in biased estimation.

Images