Grid parameterization-based remote sensing image target recognition training sample generation method

Abstract

A remote sensing image target recognition training sample generation method based on grid parameterization comprises the following steps: for a given original image, constructing a grid of a corresponding size according to the size of the given original image and triangularizing the grid to obtain an initial triangular grid M, and parameterizing the initial triangular grid M to obtain a parameterization grid G '; after unitization is carried out on the parameterized grid G ', two fixed points gamma 1 and gamma 2 on the parameterized grid G' are selected, corresponding Mobius transformation is obtained through reverse deduction of gamma 1 and gamma 2, then Mobius transformation is carried out on the parameterized grid G ', and unitization is carried out on the transformation result; and finally, according to a real-time pipeline rendering rule, taking the grid G'obtained after Mobius transformation as a parameterized coordinate, taking the original image as a texture, and pasting the texture back to the initial triangular grid M to obtain an augmented image.

Description

technical field [0001] The invention relates to a method for generating training samples for remote sensing image target recognition based on grid parameterization and Mobius transformation, and belongs to the technical fields of computational geometry and deep learning. Background technique [0002] In remote sensing image target recognition tasks, deep learning methods are usually used to detect targets such as ships, aircraft, and cars. In this process, the parameter training of the entire network often requires massive data as training samples, and the cost is relatively high. [0003] Generally, when the training data is relatively small, data augmentation is a common method to expand it and generate new samples. Its purpose is to increase the richness of the training data set and make the training data as diverse as possible, so that the training The resulting network model has stronger generalization ability. According to the timing of data processing, data augmenta...

AI Technical Summary

Problems solved by technology

In this process, the parameter training of the entire network often requires massive data as training samples, and the cost is relatively high.

Images