Semantic segmentation network model uncertainty quantification method based on evidence reasoning

Abstract

The invention provides a semantic segmentation network model uncertainty quantification method based on evidence reasoning. The method comprises the following steps: constructing an FCN network model, and training the FCN network model by using a training data set to obtain a trained FCN network model for semantic segmentation of image data; transplanting a D-S evidence theory to the trained FCN network model to obtain a reconstructed FCN network model; and inputting image data to be segmented into the reconstructed FCN network model, outputting a classification result of an image to be segmented by the FCN network model, and calculating an uncertain value of the classification result of each pixel point by using a D-S evidence theory index. Quantitative calculation of the semantic segmentation uncertainty can be effectively completed in a short time, the calculation efficiency is greatly improved, and time and resource cost are saved.

Description

technical field [0001] The invention relates to the technical field of semantic segmentation, in particular to a method for quantifying the uncertainty of a semantic segmentation network model based on evidence reasoning. Background technique [0002] Semantic segmentation is identifying what is present in an image and where (by finding all pixels belonging to it). Semantic segmentation is a typical computer vision problem that involves taking as input some raw data (e.g., planar images) and transforming them into new data with highlighted regions of interest. Semantic segmentation is classification at the pixel level, and pixels belonging to the same class are classified into one class. [0003] D-S evidence theory (Dempster-Shafer Theory of Evidence), also known as D-S theory, was the earliest work done by Dempster in using the upper and lower bound probability to solve the multi-valued mapping problem. He tried to use a probability range instead of a certain probability ...

AI Technical Summary

Problems solved by technology

Moreover, when combining the Bayesian learning method with the neural network, since the neural network itself already has a very large amount of parameters, coupled with a large number of model inferences of the Bayesian learning method, the amount of calculation is very large, and it cannot be solved in a short time. Effectively complete the inference calculation of semantic segmentation uncertainty

Images