Semantic Segmentation Using PSP Network with Attention Mechanism

Abstract

Semantic segmentation is a challenging problem in computer vision. In recent years, the performance of semantic segmentation has been considerably enhanced by employing cutting edge technique.This paper presents an advanced semantic segmentation methodology that uses the PSPNet (Pyramid Scene Parsing Net-work) architecture augmented with atrous convolution networks and a spatial attention module . The primary objective is to improve segmentation accuracy by integrating spatial attention mechanisms with the PSPNet framework, in association with atrous convolution networks. The spatial attention module selec-tively highlights pertinent spatial regions within feature maps, enhancing the ability of the model to capture intricate details crucial for precise segmentation. Experimental evaluations are carried out in two datasets: Stanford Background dataset and the Aerial Semantic Segmentation Drone dataset.This improvement underscores the efficacy of integrating spatial attention mechanisms and atrous convolution networks within the PSPNet architecture for semantic segmentation tasks, propelling advancements in the state-of-the-art performance within this domain.