{"title":"基于自适应多尺度特征融合的多任务语义分割网络","authors":"Huilin Chen, Shengsong Yang, Ting Lyu","doi":"10.1109/CCPQT56151.2022.00018","DOIUrl":null,"url":null,"abstract":"A multi-task semantic segmentation network architecture based on adaptive multi-scale feature fusion is proposed, which improves segmentation target edge details and small-scale target segmentation accuracy by combining boundary detection tasks and semantic segmentation tasks. The critical component of the architecture is the adaptive multi-scale feature fusion module, which can adaptively fuse the semantic feature information and boundary feature information of different scales, extract semantic features that contain more spatial data, and reduce the loss of spatial information of small-scale targets, thereby enhancing the network's ability to learn small-scale target features and boundary features. Experiments show that our designed network architecture can improve the segmentation accuracy of small-scale objects and optimize the edge details of segmented objects.","PeriodicalId":235893,"journal":{"name":"2022 International Conference on Computing, Communication, Perception and Quantum Technology (CCPQT)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":"{\"title\":\"Multitask Semantic Segmentation Network Using Adaptive Multiscale Feature Fusion\",\"authors\":\"Huilin Chen, Shengsong Yang, Ting Lyu\",\"doi\":\"10.1109/CCPQT56151.2022.00018\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"A multi-task semantic segmentation network architecture based on adaptive multi-scale feature fusion is proposed, which improves segmentation target edge details and small-scale target segmentation accuracy by combining boundary detection tasks and semantic segmentation tasks. The critical component of the architecture is the adaptive multi-scale feature fusion module, which can adaptively fuse the semantic feature information and boundary feature information of different scales, extract semantic features that contain more spatial data, and reduce the loss of spatial information of small-scale targets, thereby enhancing the network's ability to learn small-scale target features and boundary features. Experiments show that our designed network architecture can improve the segmentation accuracy of small-scale objects and optimize the edge details of segmented objects.\",\"PeriodicalId\":235893,\"journal\":{\"name\":\"2022 International Conference on Computing, Communication, Perception and Quantum Technology (CCPQT)\",\"volume\":\"1 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2022-08-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"1\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2022 International Conference on Computing, Communication, Perception and Quantum Technology (CCPQT)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/CCPQT56151.2022.00018\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 International Conference on Computing, Communication, Perception and Quantum Technology (CCPQT)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/CCPQT56151.2022.00018","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Multitask Semantic Segmentation Network Using Adaptive Multiscale Feature Fusion
A multi-task semantic segmentation network architecture based on adaptive multi-scale feature fusion is proposed, which improves segmentation target edge details and small-scale target segmentation accuracy by combining boundary detection tasks and semantic segmentation tasks. The critical component of the architecture is the adaptive multi-scale feature fusion module, which can adaptively fuse the semantic feature information and boundary feature information of different scales, extract semantic features that contain more spatial data, and reduce the loss of spatial information of small-scale targets, thereby enhancing the network's ability to learn small-scale target features and boundary features. Experiments show that our designed network architecture can improve the segmentation accuracy of small-scale objects and optimize the edge details of segmented objects.