Semantic segmentation method and system based on pixel rearrangement reconstruction for streetscape understanding

Abstract

The invention relates to a semantic segmentation method and system based on pixel rearrangement reconstruction for streetscape understanding, and the method comprises the steps: carrying out the preprocessing of an input image of a training set, enabling the image to be standardized, and obtaining preprocessed images with the same size; extracting general features by using a convolutional network,then obtaining dense perforated space pyramid fusion features, and extracting coding features by using the two parts of cascade connection as a coding network; selecting an intermediate layer featurefrom the convolutional network, calculating an edge feature by combining the encoding feature, taking a dense network based on a pixel rearrangement technology as a decoding network, reconstructing the image resolution, and calculating the decoding feature; performing calculating to obtain semantic segmentation loss and auxiliary supervision edge loss, and training the deep neural network by taking minimization of weighted sum loss of the semantic segmentation loss and the auxiliary supervision edge loss as a target; and performing semantic segmentation on the to-be-segmented image by using the deep neural network model, and outputting a segmentation result. The method and the system are beneficial to improving the accuracy and robustness of image semantic segmentation.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a semantic segmentation method and system for street scene understanding based on pixel rearrangement and reconstruction. Background technique [0002] Image semantic segmentation is an important branch of computer vision in the field of artificial intelligence and an important part of image understanding in machine vision. Image semantic segmentation is to accurately classify each pixel in the image to its category, making it consistent with the visual representation of the image itself, so the image semantic segmentation task is also called pixel-level image classification task. [0003] Due to the similarity between image semantic segmentation and image classification, various image classification networks are often used as the backbone of the image semantic segmentation network after removing the last fully connected layer, and are interchangeable with each ot...

AI Technical Summary

Problems solved by technology

At the same time, in the encoding process, in order to better capture the characteristics of objects of different sizes, different receptive fields and scale information are often combined, such as spatial pyramid pooling technology with holes, which cannot effectively represent the features of the point itself, and calculate different scales. Smaller scale features are not reused when feature features

At the same time, in the existing semantic segmentation methods, transposed convolution or bilinear interpolation methods are generally used to expand the features step by step in the decoding process, so the feature size is increased step by step, and the reconstruction feature cannot be effectively reconstructed. feature reuse

And in this process, shallow features are often added to optimize the decoding process, but there is no clear optimization target for shallow features, so the ability to repair details in the reconstruction process is slightly insufficient

Images