Ship Detection Based on Improved SDD Algorithm  

作　　者：Hongcheng Chu Tianhu Wang Qiannian Miao Zeran Chen Rong Wang Wenjie Li Tao Huang 

机构地区：[1]School of Electrical Information Engineering,Jiangsu University of Technology,Changzhou,Jiangsu 213001,China [2]Changzhou High-tech Technology Innovation and Entrepreneurship Service Center,Changzhou,Jiangsu 213001,China [3]Technical Development Department,CRRC Nanjing Puzhen Rolling Stock Co.,Ltd.,Nanjing,Jiangsu 211800,China

出　　处：《Instrumentation》2024年第4期35-43,共9页仪器仪表学报(英文版)

基　　金：funded by Changzhou technology project(CZ20230025);Natural science foundation of jiangsu province(BK20150247).

摘　　要：To address the issue of inadequate detection performance for small and mediumsized densely packed vessels in ship target detection,this paper proposes an improved Single Shot Multibox Detector(SSD)model to achieve more accurate detection.The algorithm redesigns the anchor boxes to fit the ship target detection dataset better and integrates the Squeeze-and-Excitation(SE)module into the Visual Geometry Group(VGG)network to enhance the channel features of the input feature maps.Additionally,the network's ability to perceive and represent important features is further enhanced by introducing the Convolutional Block Attention Module(CBAM),which is responsible for channel and spatial attention mechanisms.Finally,the feature pyramid module is employed to fuse six layers of features from the original network,thereby improving the SSD network's capability to detect small and occluded densely packed vessel targets.The experimental results show that the model's target recognition ability for fishing vessels improved from 58.07%to 65.87%;for patrol boats,the ability increased from 94.6%to 96.03%;and for inflatable boats,it rose from 72.08%to 74.93%.The overall mean Average Precision(mAP)also increased from the original model's 80.04%to 81.22%.Additionally,by clustering prior boxes,more suitable prior boxes for vessel detection were obtained,enhancing the model's perception capabilities for both large and small vessels.

关 键 词：target detection SHIP feature fusion SDD 

分 类 号：G63[文化科学—教育学]