Research On Panoptic Segmentation Network Based On Attention Mechanism

Panoptic segmentation is a technology that combines semantic segmentation and instance segmentation.It uses shared extracted features to perform semantic and instance branching tasks,and ultimately generates a global image by fusing the results of the two branches.Panoptic segmentation technology can accurately display the boundary positions of objects such as vehicles,roads,and pedestrians.Therefore,it is widely used in unmanned driving,security monitoring,virtual reality,and other fields.In order to ensure the reliability of practical applications and improve the accuracy of panoptic segmentation,this paper introduces an attention mechanism to improve the performance of panoptic segmentation networks.The main research work is as follows.Aiming at the problems of incomplete feature extraction and poor semantic and instance segmentation in panoptic segmentation networks,this paper proposes a panoptic segmentation network CPSNet based on Convolutional Block Attention Module(CBAM)and coordinate attention(CA).CPSNet is improved based on the single stage panoptic split network PSNet(panoptic swiftnet).Firstly,a CBAM module is added at the down sampling end.This module highlights the edge details of low-level features by learning the correlation between channels,and improves the segmentation effect by integrating low-level features and high-level features;Secondly,in the feature fusion section,a CA attention module is introduced to improve the feature extraction ability,provide efficient feature information for subsequent segmentation,and thereby improve the segmentation effect.The experimental results show that compared to the original network,CPSNet improves the panoptic quality by 1.1%,improving the segmentation effect of semantic objects and instance objects to a certain extent.At the same time,in view of the problem that some scale features are easily lost in panoptic segmentation networks,resulting in inaccurate panoptic segmentation and unclear edge contours,this paper proposes a panoptic segmentation network PPSNet based on Pyramid Squeeze Attention(PSA).PPSNet is improved based on an efficient panoptic segmentation network(EPSNet).In the bidirectional FPN module,a pyramid segmentation attention mechanism is introduced,combining contextual features from adjacent scales,and learning feature information from different scales to achieve interaction between different channels,there by improving the segmentation effect.The experimental results show that compared to the original network,PPSNet improves the panoptic quality by 0.8% and improves the panoptic segmentation prediction effect.In the actual application scenario of panoptic segmentation,this paper designs and implements a panoptic segmentation system based on CPSNet,which includes registration,login,upload image and panoptic segmentation functions.On the basis of achieving panoptic segmentation,it provides visualization and interaction functions,facilitating users to intuitively obtain data processing results.