Siyuan Wu, Jieyi Liu, Hongliang Luo, Zhao Nie, Hao Li, Jie Wu
{"title":"机载仪器光学检测文本区域自动选择方法","authors":"Siyuan Wu, Jieyi Liu, Hongliang Luo, Zhao Nie, Hao Li, Jie Wu","doi":"10.1109/CCIS53392.2021.9754630","DOIUrl":null,"url":null,"abstract":"For text region selection on airborne instruments scene, the character-like elements such as pointers and grids have a negative effect on text detection, and the existing text detection methods are difficult to handle it. This paper proposes a modified text detection model based on Fully Connected Neural Network and U-net, which achieved better prediction performance of fewer noise pixels, fewer wrongly predicted areas and have relatively higher spatial consistency. To further address the problem of FCN lacking spatial consistency, a method of filtering False Positives by seed anchor was proposed in post processing. The simulation result shows that the improved FCN text detection model performs better than the original Fully Connected Neural Network in both precision and recall. Furthermore, the proposed post processing method further improved precision index.","PeriodicalId":191226,"journal":{"name":"2021 IEEE 7th International Conference on Cloud Computing and Intelligent Systems (CCIS)","volume":"208 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2021-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"2","resultStr":"{\"title\":\"An Automatic Text Region Selection Method on Optical Inspection for Airborne Instrument\",\"authors\":\"Siyuan Wu, Jieyi Liu, Hongliang Luo, Zhao Nie, Hao Li, Jie Wu\",\"doi\":\"10.1109/CCIS53392.2021.9754630\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"For text region selection on airborne instruments scene, the character-like elements such as pointers and grids have a negative effect on text detection, and the existing text detection methods are difficult to handle it. This paper proposes a modified text detection model based on Fully Connected Neural Network and U-net, which achieved better prediction performance of fewer noise pixels, fewer wrongly predicted areas and have relatively higher spatial consistency. To further address the problem of FCN lacking spatial consistency, a method of filtering False Positives by seed anchor was proposed in post processing. The simulation result shows that the improved FCN text detection model performs better than the original Fully Connected Neural Network in both precision and recall. Furthermore, the proposed post processing method further improved precision index.\",\"PeriodicalId\":191226,\"journal\":{\"name\":\"2021 IEEE 7th International Conference on Cloud Computing and Intelligent Systems (CCIS)\",\"volume\":\"208 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2021-11-07\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"2\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2021 IEEE 7th International Conference on Cloud Computing and Intelligent Systems (CCIS)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/CCIS53392.2021.9754630\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2021 IEEE 7th International Conference on Cloud Computing and Intelligent Systems (CCIS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/CCIS53392.2021.9754630","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
An Automatic Text Region Selection Method on Optical Inspection for Airborne Instrument
For text region selection on airborne instruments scene, the character-like elements such as pointers and grids have a negative effect on text detection, and the existing text detection methods are difficult to handle it. This paper proposes a modified text detection model based on Fully Connected Neural Network and U-net, which achieved better prediction performance of fewer noise pixels, fewer wrongly predicted areas and have relatively higher spatial consistency. To further address the problem of FCN lacking spatial consistency, a method of filtering False Positives by seed anchor was proposed in post processing. The simulation result shows that the improved FCN text detection model performs better than the original Fully Connected Neural Network in both precision and recall. Furthermore, the proposed post processing method further improved precision index.