Mars terrain segmentation method based on double-branch input neural network

Abstract

The invention discloses a Mars terrain segmentation method based on a double-branch input neural network, and the method comprises the steps: carrying out the combined training through double branches after a sample is input into a network, one is a conventional branch which is responsible for the extraction of the information of an internal region of a target, and the other is a boundary branch which is responsible for the extraction of the edge information of the target. Information extracted by the two branches passes through a fusion module, regional features and boundary features are combined, finally a fine semantic segmentation result is output, segmentation of the Mars topography with inaccurate annotation is completed, and the aim of accurately segmenting the boundary of the Mars topography is achieved. According to the method, through introduction of the double-branch information, the accurately-labeled target area interior information and the inaccurately-labeled target boundary information are utilized at the same time, the labeling problem of inaccurate label segmentation is solved, and a large amount of time cost and labor cost can be saved.

Description

technical field [0001] The invention belongs to the field of image recognition, and relates to a Mars terrain segmentation method based on a double-branch input neural network. Background technique [0002] It is one of the development directions of my country's future deep space exploration to carry out large-scale patrol detection on the surface of extraterrestrial celestial bodies in an unmanned or manned manner. Whether it can efficiently and accurately identify unknown and complex surface environments is a key issue affecting the detection efficiency. [0003] Different from the unknown environment in the field of the earth, the extraterrestrial surface terrains such as the moon, Mars and asteroids have the characteristics of indistinct features and little prior knowledge. The traditional perception method restores the elevation map through stereo vision, and judges whether it is an obstacle according to the height of the terrain. The algorithm is complex and the featur...

AI Technical Summary

Problems solved by technology

The present invention solves the problem of inaccurate labeling of segmentation labels by introducing the information of the two branches, and utilizes the inside of the target area accurately marked and the boundary information of the target area marked inaccurately, and can save a lot of time and labor costs

Images