Super-pixel-based remote sensing image semantic segmentation method under known sample imbalance condition

Abstract

The invention provides a super-pixel-based remote sensing image semantic segmentation method under a known sample imbalance condition, and the method comprises the steps: firstly carrying out the over-segmentation preprocessing of image data, obtaining a super-pixel segmentation result, cutting a rectangular image through employing the gravity center of each super-pixel as a central point, and constructing a super-pixel rectangular image data set; training an ensemble classification model, under the condition that the known samples are unbalanced, combining the small samples into one class, carrying out primary classification, carrying out secondary classification on the small samples after the primary classification is completed, and integrating the network models of the two classifications to obtain the ensemble classification model; inputting a to-be-predicted super-pixel rectangular image data set into the integrated classification network model to obtain a super-pixel rectangularimage category; and mapping the category of the super-pixel rectangular image to the original image to obtain a semantic segmentation result. According to the method, the semantic segmentation precision of the small sample under the condition that the known sample is unbalanced can be effectively improved finally.

Description

technical field [0001] The invention relates to computer image processing and Gaofen-2 satellite remote sensing image ground object classification technology. Background technique [0002] Remote sensing image interpretation is an important part of digital image analysis, widely used in land surveying and mapping, environmental monitoring, urban construction, mineral investigation, agricultural monitoring, military command and other fields. Through the acquisition, processing and analysis of remote sensing data, people can obtain a large amount of effective information in a short time and realize the dynamic monitoring of targets. Compared with traditional manual survey work methods, remote sensing technology has the advantages of high efficiency and low cost. [0003] Due to the development of deep learning, it has become a popular direction to use deep learning methods to directly extract ground feature information to achieve ground feature information segmentation and cl...

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved by the present invention is to propose a method to solve the impact of known sample imbalance on the segmentation results by secondary classification of superpixels

Images