LIU L L, DU M M, ZHENG W,et al.Campus scene segmentation based on salient semantic collaborative[J].Journal of Henan Polytechnic University(Natural Science) ,2023,42(4):148-155.

doi:10.16186/j.cnki.1673-9787.2022070040

Received:2022/07/18

Revised:2022/12/07

Published:2023/07/25

Campus scene segmentation based on salient semantic collaborative

LIU Lanlan1,2, DU Minmin1, ZHENG Wei2, SIMA Haifeng1

1.School of Software，Henan Polytechnic University，Jiaozuo  454000，Henan，China;2.Faculty of Arts and Law，Henan Polytechnic University，Jiaozuo  454000，Henan，China

Abstract:Aimed at low segmentation accuracy and insufficient recognition of small targets，a semantic segmentation algorithm based on salience collaborative was proposed，it was an improved model of the real-time network LinkNet.Firstly，the Atrous Spatial Pyramid Pooling was introduced to enlarge receptive field.This measure was able to enrich the context semantic information，thus the segmentation ability of small targets could be improved.Secondly，collaborative learning was adopted to share feature extraction process.The processes included semantic segmentation and saliency detection.The segmentation task benefited from training of saliency by sharing the features of convolution layer，then the accuracy of this model was improved.The proposed model was trained on the Cityscapes dataset to verify the performance.Experiment results showed that the proposed algorithm improved the accuracy of various targets of scene to 67.91%，which was 8.14% higher than the original LinkNet model.

Key words:campus safety;image semantic segmentation;collaborative learning;salience semantic

基于显著性语义协同的校园道路场景分割_刘蓝蓝.pdf