摘要
针对光电图像语义分割问题,提出了一种基于编解码(Encoder-Decoder)结构和图像局部增强的分割算法。首先,采用基于互质因子的空洞空间金字塔池化(Atrous Spatial Pyramid Pooling,ASPP)模块减小多尺度空洞卷积(Atrous Convolution)引入的网格效应,提升卷积核的像素近邻信息表征能力;其次,对分割难度较大的图像局部区域,采用融合平均交并比(Mean Intersection Over Union,MIOU)和交叉信息熵的损失函数,结合权值衰减策略,提高这些局部区域的像素权重。实验结果表明,提出的改进算法能有效提升图像语义分割精度。
For semantic segmentation of optoelectric images, it proposes a segmentation algorithm based on encoder-decoder structure and local image enhancement. Firstly, the algorithm adopts an atrous spatial pyramid pooling(ASPP)module based on coprime factors to reduce the grid effect caused by multiscale convolution and improve the capability of representing pixel adjacent information of convolution kernels. Secondly, in order to improve pixel weight of areas that are difficult to segment, the algorithm combines loss function consisting of mean intersection and cross entropy with weight decay strategy. Experimental results show that the proposed algorithm can effectively improve the accuracy of image semantic segmentation.
作者
李承珊
蒋平
崔雄文
马震环
雷涛
LI Chengshan;JIANG Ping;CUI Xiongwen;MA Zhenhuan;LEI Tao(Institute of Optics and Electronics,Chinese Academy of Sciences,Chengdu 610209,CHN;University of Chinese Academy of Sciences,Beijing 100049,CHN)
出处
《半导体光电》
CAS
北大核心
2018年第6期892-897,共6页
Semiconductor Optoelectronics
基金
中国科学院青年创新促进会项目(2016336)
关键词
语义分割
空洞卷积
编解码结构
互质因子
局部增强
semantic segmentation
atrous convolution
encoder-decoder structure coprime factors
local enhancement
