融合残差连接的图像语义分割方法  

Image Semantic Segmentation Method for Fusion Residual Connection

作　　者：王龙宝[1,2] 张珞弦 张帅 徐亮 曾昕 徐淑芳[1,2] WANG Longbao;ZHANG Luoxian;ZHANG Shuai;XU Liang;ZENG Xin;XU Shufang(College of Computer and Information,Hohai University,Nanjing 210000,China;Key Laboratory of Water Big Data Technology of Ministry of Water Resources,Hohai University,Nanjing 210000,China;Power China Kunming Engineering Corporation Limited,Kunming 650000,China;Yangtze River Ecology and Environment Co.,Ltd.,Wuhan 430061,China)

机构地区：[1]河海大学计算机与信息学院,南京210000 [2]河海大学水利部水利大数据技术重点实验室,南京210000 [3]中国电建集团昆明勘测设计研究院有限公司,昆明650000 [4]长江生态环保集团有限公司,武汉430061

出　　处：《计算机测量与控制》2024年第1期157-164,共8页Computer Measurement &Control

基　　金：云南省科技厅重大科技专项计划项目(202202AF080003);长江生态环保集团有限公司科研项目(HBZB2022005)。

摘　　要：由于传统SegNet模型在采样过程中产生了大量信息损失,导致图像语义分割精度较低,为此提出了一种融合残差连接的新型编-解码器网络结构:文中引入了多残差连接策略,更为全面地保留了多尺度图像中包含的大量细节信息,降低还原降采样所带来的信息损失;为进一步加速网络训练的收敛效率,改善样本的不平衡问题,设计了一种带平衡因子的交叉熵损失函数,对正负样本不平衡现象予以针对性的优化,使得模型的训练更加高效;实验表明该方法较好地解决了语义分割中信息损失以及分割不准确的问题,与SegNet相比,本网络在Cityscapes数据集上进行精细标注的mIoU值提高了约13%。Due to the large amount of information loss generated by traditional SegNet model during the sampling process,it causes the accuracy of image semantic segmentation low.Therefore,a new encoder-decoder network structure with fusion residual connection is proposed.The multi-residual connection strategy is introduced to fully retain a large number of detailed information contained in multi-scale images,and reduce the information loss caused by sampling.In order to further accelerate the convergence efficiency of network training and improve the imbalance problem of samples,a cross-entropy loss function with balance factor is designed,and the imbalance phenomenon of positive and negative samples is emphatically optimized to train the model more efficient.Experimental results show that this method solves the problems of information loss and inaccurate segmentation in semantic segmentation,and compared with SegNet model,the fine labeling mean intersection over union(mIoU)index of the network on Cityscapes dataset is increased by about 13%.

关 键 词：语义分割 残差连接 交叉熵损失函数 SegNet模型 深度学习 

分 类 号：P391[天文地球—地球物理学]