摘要
基于内容的遥感图像检索存在着低层视觉特征与用户对图像理解的高层语义不一致、图像检索精度低以及单一的距离度量方法不能完全真实反映图像之间相似程度等问题。对此提出一种基于改进的模糊C均值聚类和卷积神经网络的遥感图像检索方法。该方法充分利用遥感图像的特性,通过Retinex算法自适应处理遥感图像噪声,运用自学习能力良好的卷积神经网络对遥感图像进行多层神经网络的监督学习,提取遥感图像特征,并运用改进的模糊C均值进行特征聚类分析。同时,将快速排序算法与距离位置权重相结合的Top-k排序算法运用到实验当中,提高遥感图像的检索精度。实验表明,该方法可以显著提高遥感图像的检索性能。
Aiming at the problems in content-based image retrieval, such as inconsistence between low-level visual features and the user′s high-level semantics for image understanding, low image retrieval accuracy, and the inability of a single distance measurement method for complete reflection of the similarity degree between images, we propose a remote sensing image retrieval method based on improved fuzzy C-means clustering and convolutional neural network （CNN）. This method makes full use of the characteristics of remote sensing images. It adaptively processes the noise of remote sensing images by using Retinex algorithm, and uses CNN to supervise the remote sensing images by multi-layer neural network to extract remote sensing image features. Besides, the modified fuzzy C-means clustering is adopted for feature clustering analysis. Meanwhile, the top-k sorting algorithm which combines the quick sorting algorithm with the distance position weights is applied to improve the retrieval accuracy of the remote sensing images. Experimental results show that this method can significantly improve the performance of remote sensing image retrieval.
作者
彭晏飞
宋晓男
訾玲玲
王伟
Peng Yanfei;Song Xiaonan;Zi Lingling;Wang Wei(School of Electronic and Information Engineering,Liaoning Technical University,Huludao,Liaoning 125105,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2018年第9期167-176,共10页
Laser & Optoelectronics Progress
基金
国家自然科学基金(61702241)
辽宁省教育厅高等学校基本科研项目(LJ2017FBL004)
辽宁省教育厅城市研究院一般项目(LJCL008)
辽宁省博士科研启动基金(201601365)
