Semantic segmentation method and system based on dynamic interpolation reconstruction for streetscape understanding

Abstract

The invention relates to a semantic segmentation method and system based on dynamic interpolation 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 characteristics by using a convolutionalnetwork, then obtaining hybrid space pyramid pooling characteristics, and extracting coding characteristics by using the two parts of cascading as a coding network; selecting an intermediate layer characteristic from the convolutional network, calculating an interpolation weight characteristic in combination with the encoding characteristic, constructing a decoding network in a dynamic interpolation mode, reconstructing an image resolution, and calculating a decoding characteristic; performing calculating to obtain edge-enhanced semantic segmentation loss, and training the deep neural networkwith the purpose of minimizing the edge-enhanced semantic segmentation loss; and performing semantic segmentation on the to-be-segmented image by using the deep neural network model, and outputting asegmentation 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 technology, in particular to a semantic segmentation method and system based on dynamic interpolation reconstruction for street view understanding. 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...

AI Technical Summary

Problems solved by technology

In the existing image semantic segmentation methods, the reconstruction of the resolution in the decoding process often uses bilinear interpolation, transposed convolution, and sub-pixel convolution to measure the size of the image. The first method is well selected Interpolate reference points, but use the physical distance of pixels as interpolation, because the semantic distance of different maps is often not the same as the physical, so it cannot effectively adapt to the situation of different images

The second is because the size of the original image is enlarged by padding zeros and then the general convolution is used for learning, so the learning efficiency is low

The last one uses the same reference point for several pixels to learn, and the selected point is not good.

There are certain problems in the above methods, which lead to the fact that large-scale features cannot be effectively learned from small-scale features with missing information during the decoding process.

Images