文献类型：期刊文献

英文题名：Automatic segmentation of golden pomfret based on fusion of multi-head self-attention and channel-attention mechanism

作者：Yu, Guoyan[1,3];Luo, Yingtong[1,2];Deng, Ruoling[1,3]

机构：[1]Guangdong Ocean Univ, Sch Mech Engn, Zhanjiang 524088, Guangdong, Peoples R China;[2]Guangdong Ocean Univ, Guangdong Marine Equipment & Mfg Engn Technol Res, Zhanjiang 524088, Guangdong, Peoples R China;[3]Southern Marine Sci & Engn Guangdong Lab Zhanjiang, Zhanjiang 524088, Guangdong, Peoples R China

年份：2022

卷号：202

外文期刊名：COMPUTERS AND ELECTRONICS IN AGRICULTURE

收录：SCI-EXPANDED(收录号：WOS:000863429900004)、、EI(收录号：20223912812352)、Scopus(收录号：2-s2.0-85138464190)、WOS

基金：The work was financially supported by the Guangdong Inter-regional Collaborative Fund (No. 2019B1515120017) , Guangdong Special Project of Ocean Economic Development (No.011Z21001), Zhanjiang project of Innovation and Entrepreneurship Team "Pilot Program" (No.2020LHJH003) , and Program for scientific research start-up funds of Guangdong Ocean University (No. 060302062106) .

语种：英文

外文关键词：Golden pomfret; Instance segmentation; Multi -Head self -attention; Channel -attention

外文摘要：In precision fishery, the accurate segmentation of each golden pomfret image from the background is an important step to obtain golden pomfret information in real time. However, in complex sea conditions and highly occluded scenes, traditional segmentation methods are still challenging to segment golden pomfret with high speed and high accuracy. In this study, a novel model named SE-TongNet, which fuses multi-head self-attention mechanism and channel mechanism based on Mask R-CNN framework, is proposed for automatic segmentation of golden pomfret. Among them, the multi-head self-attention module creates the sparse attention map for enhancing the fine-grained features of golden pomfret, which can better meet the requirements of real-time detection and segmentation. Also, a novel channel attention mechanism is embedded to filter the redundant information of some channels, thereby optimizing the model. Overall, the SE-TongNet model enables robust learning with multi-policy understanding of high-level semantics in coupled noise scenarios and improves computational efficiency. The test results show that compared with other state-of-the-art networks, the improved method can accurately and effectively segment golden pomfret, with mAP and segmentation rate reaching 82.55 and 5.31fps, respectively. Furthermore, the performance of the SE-TongNet is robust and practical in four major scenarios.