4.5 Article

An enhancement model based on dense atrous and inception convolution for image semantic segmentation

Journal

APPLIED INTELLIGENCE
Volume 53, Issue 5, Pages 5519-5531

Publisher

SPRINGER
DOI: 10.1007/s10489-022-03448-w

Keywords

Image semantic segmentation; Dense convolution; Atrous convolution; Inception convolution; Full convolution neural network

Ask authors/readers for more resources

This paper proposes a novel semantic segmentation model called Ince-DResAsppNet based on dense convoluted separation convolution. The model aims to reduce semantic information loss and enhance detailed information in order to improve pixel-level semantic understanding. Experimental results demonstrate that our model outperforms existing semantic segmentation models in terms of segmentation accuracy on the PASCAL VOC 2012 and CityScapes datasets.
The goal of semantic segmentation is to classify each pixel in the image, so as to segment out the specific contour of the target. Most previous semantic segmentation models cannot generate enough semantic information for each pixel to understand the content of complex scenes. In this paper, we propose a novel semantic segmentation model Ince-DResAsppNet based on dense convoluted separation convolution. Unlike the previous model, our model revolves around reducing semantic information loss and enhancing detailed information. In the feature extraction part of the model, the idea of Dense and Ince is introduced to expand the number of channels on the basis of feature reuse. In the feature fusion part, Dense and Atrous's idea of dense dilated based on coprime factors is introduced, combined with multi-scale feature information to expand the receptive field and collect more dense pixels. Experiments conducted on the dataset PASCAL VOC 2012 and the CityScapes dataset show that our method performs better than the existing semantic segmentation model. Our model achieves 83.3% and 78.1% segmentation accuracy on the mIoU indicator, which surpasses many classical semantic segmentation models.

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

4.5
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

No Data Available
No Data Available