基于改进SOLOv2的集装箱货箱图像实例分割  

Image segmentation of container based on improved SOLOv2

作　　者：苏铁明[1] 梁琛 徐志祥[1] 李鹏博 王宣平 刘玮 SU Tie-ming;LIANG Chen;XU Zhi-xiang;LI Peng-bo;WANG Xuan-ping;LIU Wei(School of Mechanical Engineering,Dalian University of Technology,Dalian 116081,Liaoning Province,China;Jiuyi Aerospace Technology(Dalian)Co.,Ltd.,Dalian 116085,Liaoning Province,China)

机构地区：[1]大连理工大学机械工程学院,辽宁大连116081 [2]久亿航宇科技(大连)有限公司,辽宁大连116085

出　　处：《信息技术》2024年第11期112-119,共8页Information Technology

基　　金：大连市揭榜挂帅科技攻关项目(2022YJ11SN00103)。

摘　　要：针对集装箱货箱图像实例分割存在分割准确度不高的问题,在SOLOv2算法模型基础上,将主干网络中的残差网络替换为Swin Transformer网络,增强提取货箱特征信息的能力;将随机梯度下降优化器替换为AdamW优化器,加快模型的收敛速度;引入复制粘贴数据增强方法,混合粘贴实例对象,增加训练数据;采集不同摆放位置的货箱图像制作成数据集,利用添加噪声等图像增强方法扩充数据集,提高SOLOv2的分割性能。在集装箱货箱数据集上进行测试,结果表明改进SOLOv2相比SOLOv2的掩膜平均精度提高了3.0%,有效提高了集装箱货箱图像实例分割精度。To improve the low segmentation accuracy of container image instances,based on SOLOv2 algorithm model,the residual network in trunk network is replaced with Swin Transformer network to enhance the ability of extracting container feature information.Stochastic gradient descent optimizer is replaced by AdamW optimizer to accelerate the convergence rate of the model.The copy and paste data enhancement method is introduced,and the instance object is mixed and pasted to increase the training data.The images of cargo boxes at different locations are collected and made into data sets.The data sets are expanded by adding noise and other image enhancement methods to improve the segmentation performance of SOLOv2.The test is carried out on the container data set,and the results show that the improved SOLOv2 improved the average accuracy of mask by 3.0%compared with SOLOv2,which effectively improved the segmentation accuracy of cargo boxes image instances.

关 键 词：SOLOv2 Swin Transformer AdamW 数据增强 集装箱货箱 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]