文献类型：期刊文献

英文题名：Discrete-Time Noise-Suppression Neural Dynamics for Optical Remote Sensing Image Extraction

作者：Xiong, Yizhen[1];Wang, Difeng[2];Fu, Dongyang[1];Wang, Yan[1]

机构：[1]Guangdong Ocean Univ, Sch Elect & Informat Engn, Zhanjiang 524088, Guangdong, Peoples R China;[2]Minist Nat Resources, Inst Oceanog 2, State Key Lab Satellite Ocean Environm Dynam, Hangzhou 310012, Zhejiang, Peoples R China

年份：2023

卷号：11

起止页码：92111

外文期刊名：IEEE ACCESS

收录：SCI-EXPANDED(收录号：WOS:001060323200001)、、EI(收录号：20233514640375)、Scopus(收录号：2-s2.0-85168670592)、WOS

基金：This work was supported in part by the Key Projects of the Guangdong Education Department under Contract 2019KZDXM019, in part by the Fund of Southern Marine Science and Engineering Guangdong Laboratory (Zhanjiang) under Contract ZJW-2019-08, in part by the High-Level Marine Discipline Team Project of Guangdong Ocean University under Contract 002026002009, in part by the Guangdong Graduate Academic Forum Project under Contract 230420003, in part by the 2019 First Class Discipline Construction Platform Project of Guangdong Ocean University under Contract 231419026, in part by the National Natural Science Foundation of China under Contract 41476157, and in part by the National Key Research and Development Program of China under Grant 2018YFB0505005.

语种：英文

外文关键词：Target object extraction; discrete-time noise-suppression neural dynamics (DTNSND) model; constrained energy minimization (CEM) scheme; noise-suppression

外文摘要：Optical remote sensing is an important method of observing objects over large areas. Naturally, it is essential to extract the target from optical remote sensing images. Most existing methods, such as thresholding methods and texture analysis-based methods, have some limitations. Additionally, most methods are generally not robust to noise, which tends to affect extraction results to some extent. Thus, how to extract the target object from optical remote sensing images conveniently and robustly is a challenge. To make up for the shortcomings of most methods, a constrained energy minimization (CEM) scheme is applied to extract the target object. Then, a discrete-time noise-suppression neural dynamics (DTNSND) model with an error-accumulation term is proposed to aid the CEM scheme for extracting the target object, which restrains the effects of noises in the extraction process. Theoretical analyses demonstrate that the DTNSND model suppresses noise in diverse noisy environments. Furthermore, numerical simulations are provided to illustrate that the maximal steady-state residual error generated by the DTNSND model is markedly lower than those of comparative algorithms. Finally, extraction experiments, using an optical remote sensing image of the Arctic sea ice as an experimental material, are executed in zero noise and random noise environments, respectively. Comparative results confirm that the DTNSND model is able to extract the remote sensing image stably and accurately in noisy environments, further demonstrating the feasibility of the DTNSND model in practice.