A Discretization Method and System for Remote Sensing Image Features Based on Information Entropy

Abstract

The invention relates to a method and system for discretizing remote sensing image features based on information entropy. The method includes the steps of: establishing an image information decision table based on pixel values ​​and categories; calculating interval information entropy based on an interval separation model based on information entropy The information entropy is divided into intervals; the image information decision table is discretized, and the indistinguishable relationship difference degree of the image information table before and after the discretization is calculated respectively; the result of the discretization is evaluated, and the optimal discretization scheme is selected . The method and system of the invention can reduce the number of discretization intervals, avoid the redundancy of interval information, and alleviate the problem of discretization data distortion.

Description

technical field [0001] The invention relates to the technical field of remote sensing image feature extraction, in particular to a method and system for discretizing remote sensing image features based on information entropy. Background technique [0002] Environmental monitoring is an important part of marine information construction, which provides a strong scientific basis for marine environmental protection and management, resource development and utilization. Ocean satellites effectively make up for the shortcomings of traditional ocean observation methods. Based on the continuous observation of the ocean based on a variety of remote sensors, humans have greatly deepened their understanding of the ocean. and environmental protection and many other fields play an important role. To analyze and process the collected remote sensing image data, it is necessary to extract its features first, and these image features are often continuous data. However, most of the current k...

AI Technical Summary

Problems solved by technology

[0004] At this stage, there are still relatively few mature unsupervised methods. In the absence of class information, it is difficult to obtain good discretization results, and the discretization results are also difficult to measure.

In the supervised discretization algorithm, the univariate supervised discretization algorithm does not take into account the interdependence between attributes, but only develops from the unsupervised discretization mode to consider the one-way dependence of an attribute to a category in isolation or is the interdependence of attributes and categories

When such a discretization algorithm is applied to the classification process, a large number of redundant breakpoints between attributes and the absence of necessary breakpoints will lead to the failure of the overall process

Although the existing multivariable supervised discretization algorithm (Multi Variable Supervised Discretization) has made up for the shortcomings of the above-mentioned univariate supervised discretization algorithm to a certain extent, it has not yet achieved the goal of eliminating redundant breakpoints while maintaining Indistinguishable relation of decision table

When selecting breakpoints, it does not take into account the mutual exclusion of breakpoints between attributes and within attributes, which is obviously not suitable for processing remote sensing image data with multiple features.

Images