A Road Scene Semantic Segmentation Method Effectively Fusion of Neural Network Features

Abstract

The invention discloses a road scene semantic segmentation method that effectively integrates neural network features. It constructs a convolutional neural network in the training stage, which includes an input layer, a hidden layer, and an output layer. The hidden layer includes three neural network blocks. Spatial feature extraction channel, background feature extraction channel composed of 5 neural network blocks, and feature fusion channel composed of the fifth fusion block; each original road scene image in the training set is input into the convolutional neural network for training , to obtain 12 semantic segmentation prediction maps corresponding to each original road scene image; by calculating the set of 12 semantic segmentation prediction maps corresponding to each original road scene image and processing the 12 corresponding real semantic segmentation images Convolutional neural network training model is obtained by one-hot encoding the loss function value between sets of images; in the test phase, the convolutional neural network training model is used for prediction; the advantage is high segmentation accuracy and strong robustness.

Description

technical field [0001] The invention relates to a semantic segmentation method, in particular to a road scene semantic segmentation method which effectively integrates neural network features. Background technique [0002] Semantic segmentation is a fundamental technique for many computer vision applications, such as scene understanding, autonomous driving, etc. With the development of Convolutional Neural Networks, especially Fully Convolutional Neural Networks (FCNs), many promising results have been achieved on benchmarks. The fully convolutional neural network has a typical encoder-decoder structure, semantic information is first embedded into the feature map through the encoder, and the decoder is responsible for generating segmentation results. Typically, the encoder extracts image features through a pre-trained convolutional model, and the decoder contains multiple upsampling components to recover resolution. Although the encoder's most important feature maps may be...

AI Technical Summary

Problems solved by technology

But due to the difference in semantic level and spatial resolution, simple fusion of low-level and high-level features may not be effective

Images