首页 | 官方网站   微博 | 高级检索  
     

基于深度学习的RGBD图像协同显著目标检测
引用本文:周晓飞,郭舒瑶,温洪发,刘炳涛,李世锋,张继勇,颜成钢.基于深度学习的RGBD图像协同显著目标检测[J].信号处理,2022,38(6):1213-1221.
作者姓名:周晓飞  郭舒瑶  温洪发  刘炳涛  李世锋  张继勇  颜成钢
作者单位:1.杭州电子科技大学自动化学院, 浙江 杭州 310018
基金项目:国家重点研发项目2020YFB1406604国家自然科学基金61901145浙江省自然科学基金LR17F030006杭电-中电大数据技术工程研究中心KYH063120009
摘    要:本文旨在研究一种基于深度学习的RGBD图像协同显著目标检测模型。首先,本文构建了多分支的编码器结构,有效地提取RGBD图像的深层卷积特征;然后,使用多模态特征融合模块充分融合来自编码器的深层特征;最后,通过基于残差基本块的解码器来预测得到显著性图。此外,本文以深层次监督的方式对整个网络进行约束优化。在两个公开数据集上的测试结果表明,所提模型在预测精度上优于当前6种主流模型,这其中我们的显著性图呈现出更精确的边缘细节。 

关 键 词:协同显著目标    RGBD图像    深度学习
收稿时间:2021-10-09

Deep Learning-based Co-salient Object Detection on RGBD Images
Affiliation:1.School of Automation,Hangzhou Dianzi University, Hangzhou, Zhejiang 310018, China2.School of Computer Science, Hangzhou Dianzi University, Hangzhou, Zhejiang 310018, China3.China Electric Data Services Co. Ltd., Beijing 100088, China
Abstract:? ?This paper aims to propose a co-salient object detection model on RGBD images based on deep learning algorithm. Firstly, this paper constructs a multi-stream encoder structure which can be effectively employed to extract deep convolution features of RGBD images. Then, a multi-modal feature fusion module is used to sufficiently integrate the deep features from the encoder. Finally, a decoder equipped with the residual connection and deep supervision is designed to generate saliency maps. The experimental results on two public datasets show that the performance of our model is superior to the six state-of-the-art models, where the saliency map of our model presents more precise boundary details. 
Keywords:
点击此处可从《信号处理》浏览原始摘要信息
点击此处可从《信号处理》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司    京ICP备09084417号-23

京公网安备 11010802026262号