Semantic image segmentation based on convolution neural network

Abstract

The invention discloses a semantic image segmentation method of a convolution neural network, which realizes semantic image segmentation by combining partial layered feature fusion and all-connected conditional random field processing in a network structure. The method comprises the following steps: a network architecture model is constructed to be composed of a convolution layer, a pooling layer,an upsampling layer and a loss function, wherein the step size of each layer in the model, the convolution core size and the number of output characteristic maps are customized according to specifications; the information of a shallow layer and a deep layer in the network architecture is fused, and the fused features are upsampled to the resolution of the original image; a full-connected conditional random field is used to segment the feature image. The application of the scheme of the invention fully utilizes the information of each layer and improves the final accuracy rate through the feature fusion of different layers; combined with all-connected CRF, the output of the network is post-processed, and the relationship between pixels is processed, which makes the result of image segmentation more accurate and smooth.

Description

technical field [0001] The invention belongs to the field of semantic image segmentation, and specifically refers to the implementation of semantic image segmentation by using a deep learning method. Background technique [0002] With the continuous breakthrough of deep learning, it has been widely used in various fields, such as computer vision, speech recognition, natural language processing, etc. The proposal of convolutional neural network (CNN) has made deep learning a hot word to some extent, and the proposal of fully convolutional neural network (FCN) has made a great breakthrough in semantic image segmentation. Semantic image segmentation can It is said to be the cornerstone technology of image understanding, and it plays a pivotal role in applications such as autonomous driving and drones. As we all know, an image is composed of many pixels, and semantic image segmentation, as the name implies, is to classify each pixel in the picture, that is, to realize the posit...

AI Technical Summary

Problems solved by technology

[0004] (1) To a certain extent, the information between the upper and lower layers is not well utilized, resulting in the loss of information;

[0005] (2) The connection between pixels is not considered

Images