Integrated Attention Mechanism and Edge-guided Semantic Segmentation MA-PSPNet Model for Construction Site Scenes
CHEN Hui
WANG Tao
Abstract：To enhance the semantic segmentation accuracy of safety helmets in complex scenarios and address issues involving blurred edges,poor segmentation of small objects,and multi-scale variations,a multi-scale attention pyramid scene parsing network(MA-PSPNet)model for construction site scenes integrated with attention mechanism and edge guidance was developed.In the proposed architecture,a multi-scale convolutional attention(MSCA)module was embedded within the feature extraction backbone network of the model to enhance feature representation in critical regions.An edge-guided attention(EGA)module was incorporated subsequently to the second-stage feature extraction backbone network to refine boundary identification capabilities.Furthermore,the pyramid pooling structure of the pyramid scene parsing network(PSPNet)was replaced by an atrous spatial pyramid pooling module to strengthen multi-scale adaptation.The results showed that the mean intersection over union of MA-PSPNet model was 83.28%,with an improvement of 9.13 percentage points compared to the original PSPNet model.The pixel accuracy and mean pixel accuracy were quantified at 95.62%and 88.74%,respectively.MA-PSPNet model could enhance effectively safety helmet segmentation precision and boundary awareness within complex industrial environments and had good practicality.
Keywords：safety helmet segmentationsemantic segmentationattention mechanismedge guidancemulti-scale enhancement
Publication Date：2025-09-20
Online Publishing Date：2025-09-24(First online date of this platform, not the publication date of the document)
Pages：6( 357-361,369 )