{"title":"The Improved CenterNet for Ship Detection in Scale-Varying Images","authors":"Xujia Hou, Feihu Zhang","doi":"10.1109/IAI53119.2021.9619209","DOIUrl":null,"url":null,"abstract":"Recently, deep learning for object detection has been widely used in face recognition, traffic detection, and other fields. However, due to the dataset limitation and the target scale variation issues, ship detection is not as perfect as in other fields. To address such issues, in this paper, an anchor-free detection method is proposed in framework of CenterNet. By improving the network structure and using the unique activation function, the detection issues are successfully solved. Experimental results show that the improved CenterNet method is 5.6% higher in mAP in contrast to the state-of-the-art, and the FPS is increased by 4, respectively.","PeriodicalId":106675,"journal":{"name":"2021 3rd International Conference on Industrial Artificial Intelligence (IAI)","volume":"89 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2021-11-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2021 3rd International Conference on Industrial Artificial Intelligence (IAI)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/IAI53119.2021.9619209","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Recently, deep learning for object detection has been widely used in face recognition, traffic detection, and other fields. However, due to the dataset limitation and the target scale variation issues, ship detection is not as perfect as in other fields. To address such issues, in this paper, an anchor-free detection method is proposed in framework of CenterNet. By improving the network structure and using the unique activation function, the detection issues are successfully solved. Experimental results show that the improved CenterNet method is 5.6% higher in mAP in contrast to the state-of-the-art, and the FPS is increased by 4, respectively.